Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Nova Southeastern UniversityNSUWorks
HCNSO Student Theses and Dissertations HCNSO Student Work
4-29-2016
Systematic Patterning of Sediments in FrenchPolynesian Coral Reef SystemsAndrew CalhounNova Southeastern University, [email protected]
Follow this and additional works at: https://nsuworks.nova.edu/occ_stuetd
Part of the Geology Commons, Marine Biology Commons, and the Sedimentology Commons
Share Feedback About This Item
This Thesis is brought to you by the HCNSO Student Work at NSUWorks. It has been accepted for inclusion in HCNSO Student Theses andDissertations by an authorized administrator of NSUWorks. For more information, please contact [email protected].
NSUWorks CitationAndrew Calhoun. 2016. Systematic Patterning of Sediments in French Polynesian Coral Reef Systems. Master's thesis. Nova SoutheasternUniversity. Retrieved from NSUWorks, . (406)https://nsuworks.nova.edu/occ_stuetd/406.
HALMOS COLLEGE OF NATURAL SCIENCES AND OCEANOGRAPHY
Systematic Patterning of Sediments in French Polynesian Coral Reef
Systems
By
Andrew Calhoun
Submitted to the Faculty of
Halmos College of Natural Sciences and Oceanography
in partial fulfillment of the requirements for
the degree of Master of Science with a specialty in:
Marine Biology
Nova Southeastern University
April, 2016
Thesis of
Andrew Calhoun
Submitted in Partial Fulfillment of the Requirements for the Degree of
Masters of Science:
Marine Biology
Andrew Calhoun
Nova Southeastern University
Halmos College of Natural Sciences and Oceanography
April, 2016
Approved:
Thesis Committee
Major Professor :______________________________
Sam Purkis, Ph.D.
Committee Member :___________________________
Paul (Mitch) Harris, Ph.D.
Committee Member :___________________________
Bernhard Riegl, Ph.D.
Table of Contents
Acknowledgements ........................................................................................................................ 5
Abstract ........................................................................................................................................... 6
Key Words ...................................................................................................................................... 6
1. Introduction ................................................................................................................................ 7
1.1 Objectives ............................................................................................................................. 10
2. Regional Setting and Geomorphology of Raivavae, Tubuai, and Bora Bora ..................... 11
3. Methods ..................................................................................................................................... 15
3.1 Sediment Sample Collection ............................................................................................... 15
3.2 Granularmetric Analysis of the Sediment Samples ........................................................... 17
3.3 Petrographic Analysis of the Sediment Samples ................................................................ 19
3.4 Delineating the Platform Margin and Platform Interior from Satellite Imagery ............ 21
3.5 Calculating Relative Distance of a Sediment Sample from the Reef Rim ........................ 22
4 Statistical Methods .................................................................................................................... 23
4.1 Formulating Models to Predict Water Depth and RDRR .................................................. 23
4.2 Applying Models to the Study Sites .................................................................................... 25
4.3 Assessing the Accuracy of the Predictive Models .............................................................. 25
4.4 Applying Raivavae Models to Tubuai and Bora Bora ....................................................... 26
4.5 Linear Discriminant Analysis ............................................................................................. 26
5. Results ....................................................................................................................................... 27
5.1 Sedimentary Properties ....................................................................................................... 27
5.1.1 Raivavae and Tubuai ................................................................................................... 27
5.1.2 Bora Bora ..................................................................................................................... 29
5.2 Sedimentary Properties Showed Mixed Performance for Predicting Water Depth ......... 32
5.2.1 Raivavae and Tubuai Models ...................................................................................... 32
5.2.2 Raivavae and Bora Bora Models ................................................................................. 34
5.3 Sedimentary Properties Show Moderate Accuracy for Predicting Relative Distance from
the Reef Rim .............................................................................................................................. 36
5.3.1 Raivavae and Tubuai Models ...................................................................................... 36
5.3.2 Raivavae and Bora Bora Models ................................................................................. 38
5.4 Sedimentary Properties Can’t Reliably Differentiate Between Platform Margin and
Platform Interior ....................................................................................................................... 40
5.4.1 Raivavae Linear Discriminant Analysis Model .......................................................... 40
5.4.2 Tubuai and Bora Bora Linear Discriminant Analysis Models .................................. 43
7. Discussion ................................................................................................................................. 45
7.1 Apparent Local Environmental Effects .............................................................................. 46
7.2 Abundance of Coral Fragments: Indicator of Distance from Reef Rim with Applications
to the Rock Record .................................................................................................................... 51
8. Conclusion ................................................................................................................................ 52
9. References ................................................................................................................................. 54
Appendix A: Photolog of Faunal Grain Types .......................................................................... 61
Appendix B: Referenced Tables ................................................................................................. 67
Acknowledgements
I would like to thank my principle advisor, Dr. Sam Purkis for his guidance and
encouragement throughout this thesis project. I have learned many valuable skills while
working with Dr. Purkis that will permeate into my professional career. Dr. Purkis has
played an integral part in preparing me for and guiding me into my career path.
I would also like to thank Jeremy Kerr for taking the time to share his knowledge
of statistical analysis with me. Jeremy was a remarkable mentor who helped me
understand the statistical analysis used in my thesis and explained the computer code that
ran the analysis.
Thank you to my committee members, Dr. Bernhard Riegl and Dr. Mitch Harris
for their expert advice throughout this thesis project.
Thank you to the Living Oceans Foundation for providing the means to collect
samples for this thesis project.
Abstract
Through a discipline termed “comparative sedimentology”, modern carbonate
depositional environments have been used extensively as analogs to aid in the
interpretation of equivalent fossil systems. Using field samples, GIS and remote sensing
data for three isolated carbonate platforms in the Pacific, this thesis seeks to examine
relationships between grain texture and grain type and their environment of deposition.
The motivation is to highlight relationships that have the potential to better understand
facies relations on carbonate platforms, and thereby reduce uncertainty and increase
accuracy of subsurface exploration. The results of this study show that on Raivavae,
Tubuai, and Bora Bora: French Polynesia grain texture and type of collected sediment
samples could be used to predict water depth and relative distance lagoonward from the
reef rim with ≥ 73% and ≥ 67% accuracy, respectively. The predictive relationships;
however, were largely site specific. The exception being that the same relationship
between water depth and the abundance of mud and coral could be used on both
Raivavae (accuracy = 81%) and Tubuai (accuracy = 78%). Additionally, the abundance
of coral and Halimeda in sediment samples were able to classify samples as belonging to
either the platform margin or platform interior environments on Raivavae, Tubuai, and
Bora Bora with 75%, 65%, and 65% accuracy, respectively. Overall, the results of this
study suggest that the abundance of coral holds potential to be utilized as a proxy for
distance from the reef rim on modern and ancient isolated carbonate platforms dating
back to the Miocene geological epoch.
Key Words: Isolated carbonate platform, Pacific Ocean, Petrographic
analysis, Geostatistical modeling
1. Introduction
Given the propensity of fossil reefs and their associated detritus to form excellent
water aquifers and hydrocarbon reservoirs, considerable effort has been dedicated to
understanding their anatomy, scales of accumulation and petrography. The concept of
“comparative sedimentology”, whereby facies (i.e. distinct rock or sediment bodies)
patterns are compared within and between geologic periods, has received particular
attention. In this vein the use of modern carbonate depositional environments as analogs
to ancient ones has risen to the fore. Through examination of three modern isolated
carbonate platforms in the Pacific, this thesis will develop and test a comparative
sedimentologic approach that might be utilized to more accurately interpret modern
platform facies relations and ancient subsurface carbonate stratigraphies.
The complexity of subsurface carbonate systems hinders detailed direct
characterization of their three dimensional anatomy. Information on their internal
properties is gathered from wells or outcrops. Most of these observations are along a
vertical profile and do not provide information in the lateral, which is problematic for
geostatistical modeling. While seismic data and horizontal wells mitigate this bias to
some extent, the lateral dimension of a buried system often remains vastly undersampled.
Modern analogs allow for examination of lateral trends in carbonate depositional
systems, and remote sensing coupled with ground-truth information has been used
extensively to this end (Purkis et al., 2007; Kaczmarek et al., 2010; Rankey & Reeder,
2010; Harris et al., 2011; Purkis et al., 2012a; Purkis et al., 2012b; Madden et al., 2013;
Harris et al., 2014b; Purkis et al., 2014).
Marine shallow water carbonate depositional systems have long served as modern
analogs to fossil systems. The internal precipitation of calcium carbonate has been a
common-life strategy for marine organisms since the Proterozoic, as has their
construction of carbonate platforms. As favored by modern colonial scleractinian corals,
reef forming organisms have typically adopted a niche in shallow well-lit tropical marine
waters where they are capable of building vast carbonate edifices, or platforms, that
persist into the rock record (Purkis et al., 2015a). Tucker and Wright (2009) report five
broad categories of carbonate platforms: rimmed shelves, ramps, epeiric platforms,
drowned platforms, and isolated platforms. This study concentrates on isolated platforms
(i.e. shallow water carbonate accumulations surrounded by deep water).
Isolated platforms are situated in the deep ocean. Consequently, their margins are
typically subjected to strong prevailing winds, swells and storm patterns, with the
exposure regime around the platform dictated by its orientation with regard to prevailing
open ocean hydrodynamics. Isolated platforms have a high-energy windward margin,
typically unsheltered and therefore subjected to long-period open ocean swells.
Occasionally, windward margins can be somewhat sheltered from the influence of
incident swell by neighboring platforms or antecedent topography (i.e. elevation and
relief of the Pleistocene subsurface). Windward margins are typically characterized by
reefs, rubble and coarse grained carbonate detritus (Hine et al., 1981). Finer sediments,
such as fine sands and muds, are swiftly winnowed and transported from the windward
margin towards the lower-energy platform-interior where they accumulate (Tucker &
Wright, 2009; Purkis & Harris, 2016). Depending on the influence of currents in the
platform-interior, fines might accumulate in thick deposits, else migrate further to the
leeward margin, and ultimately may be lost down the platform flanks to form
periplatform sediment wedges (Hine et al., 1981; Eberli, 1989; Wilber et al., 1990; Rees
et al., 2005; Harris et al., 2011; Purkis et al., 2012b; Betzler et al., 2014; Harris et al.,
2014b). In reality, of course, the distribution of sediment is highly variable between and
within platforms because of their complex topography and architecture that depart
radically from text-book conceptual models (e.g. Purkis et al. 2015). Accounting for and
understanding this variation has driven research on isolated platforms throughout the
world, including the Bahamas (Illing, 1954; Imbrie & Purdy, 1962; Harris, 1979, 1983;
Reijmer et al., 2012; Harris et al., 2014b; Purkis & Harris, 2016), Caribbean Sea
(Triffleman et al., 1992; Gischler & Zingeler, 2002), and Indo-Pacific (Weber &
Woodhead, 1972; Yamano et al., 2002; Purkis et al., 2005; Gischler, 2006; Riegl et al.,
2007).
The goals of recent studies have been to better understand the makeup and
geometry of carbonate facies within and among these platforms, the constraining factors
that influence these products, and how this knowledge can be applied to create and
understand a better geostatistical model (Harris & Vlaswinkel, 2008; Rankey et al., 2009;
Rankey & Reeder, 2010; Harris et al., 2011; Rankey & Reeder, 2011; Rankey & Garza-
Pérez, 2012; Wasserman & Rankey, 2014). Specifically, a study by Rankey et al. (2011)
on two Pacific atolls, Aitutaki & Maupiti, quantitatively confirmed the qualitative
understanding that sediment texture and type are, to some degree, correlated with water
depth and distance from the platform margin. The authors reported a significant positive
correlation between the abundance of mud and fine sand with increasing water depth and
distance lagoonward from the platform margin, on both atolls. This same positive trend
was noted for the abundance of non-skeletal grains, while a negative correlation was
reported for the abundance of coral fragments. Similarly, Wasserman and Rankey (2014)
found an inverse relationship between coral and coralline algae grains and increasing
distance from the platform margin.
The results from the aforementioned studies are reminiscent of a classic study by
Ginsburg (1956). In this study, the author illustrated how the relative abundances of five
major faunal grain types (Halimeda, coralline algae, coral, Foraminifera (foram), and
mollusk) showed consistent variation along transects spanning three reef sub-
environments (fore reef, outer reef arc, and back reef) of the South Florida reef tract.
Much like the findings of Rankey et al. (2011) and Wasserman and Rankey (2014),
Ginsburg (1956) found that coral and coralline algae reached maximum abundance
proximal to the outer reef arc and decreased in abundance towards the back reef. The
opposite trend was observed for Halimeda. Since the fundamental work of Ginsburg
(1956), current studies, including this study, continue to evaluate the prevalence of these
trends on a global basis.
Another question that has been thoroughly studied and debated in comparative
carbonate sedimentology is the degree to which shallow water marine carbonate facies be
linked to water depth (Rankey, 2004; Bosence, 2008; Purkis et al., 2012a; Harris et al.,
2014b; Purkis et al., 2015b)? The answer of which could be used to reconstruct
paleodepths and paloenvironments from the textural and faunal grain type composition of
carbonate rocks (Schlager, 2007). Studying a 2 km2 site around Tavernier Key, FL within
a depth range of 0 – 3 m, Bosence (2008) showed that the facies: off-mound Thalassia,
Porites, and Spongites all exhibited maximum facies abundance at distinct depth zones.
Conversely, for a 400 km2 study site off Key Largo, FL within a depth range of 0 – 9 m,
Rankey (2004) reported that the facies: rudstone, grainstone, wackestone, shelf margin
reef, and patch reef were not partitioned by depth. Likewise, Harris et al. (2014b) showed
that on the Great Bahama Bank (> 100,000 km2) the facies: rudstone, high-energy-
grainstone, grainstone, mud-poor-packstone, mud-rich-packstone, wackestone, mudstone,
and land span considerable depth ranges from 0 – 16 m. Here wackestone was the most
depth-constrained facies with a depth range of 0 – 6 m. In Saipan lagoon (65 km2) within
a depth range of 0 – 30 m, Purkis and Vlaswinkel (2012) showed mixed results where the
facies: bioclastic packstone, branching framestone, mixed siliclastic skeletal grainstone,
and wackestone were constrained by narrow depth ranges. Whereas the facies: massive
framestone, boundstone skeletal grainstone, hardground, and siliclastic grainstone
occupied much wider depth ranges. However, from a 6,000 km2 study site in the Red Sea
with a depth range of 0 – 40 m, Purkis et al. (2015) showed that none of the four
observed facies (wackestone, grainstone, boundstone, rudstone) were constrained by
water depth. The variable results from these examples highlights the need to continue to
examine the relationship between water depth and facies texture and type on other
carbonate platforms.
1.1 Objectives
The goal of this thesis are to evaluate how well grain texture (i.e. grain size and
sorting) and faunal grain type of collected sediment samples can be used to estimate the
position on three French Polynesian isolated platform tops: Raivavae, Tubuai, and Bora
Bora. Position was recorded as three separate measures: water depth, relative distance
lagoonward from the reef rim, and environment of deposition (platform margin or
platform interior). The premise being that if sedimentologic properties of a stratigraphy
vary systematically with the position on the platform top of a modern isolated carbonate
platform, the same relationships might conceivably hold for fossil subsurface systems
which are notoriously difficult to understand from sparse borings, wells and outcrops.
Raivavae was treated as a developmental site to formulate and test a suite of
mathematical models on their efficacy to predict position on the platform top based on
grain texture and faunal grain type of collected sediment samples. These models were
then deployed using collected data from Tubuai and published data from Bora Bora
(Gischler, 2011) in order to blind-test their efficacy. The best predictive model was
statistically selected and examined for potential utility for work in the subsurface. The
undertaking of this study involved: 1) granularmetric and petrographic analysis of surface
sediment samples previously collected from Raivavae and Tubuai, 2) investigation into
lateral trends in grain texture and faunal grain type with the use of remote sensing and
GIS and 3), high-level geostatistical modelling.
2. Regional Setting and Geomorphology of Raivavae, Tubuai, and Bora
Bora
Raivavae, Tubuai, and Bora Bora are part of the French Polynesia archipelago
(Fig. 1A). Five smaller archipelagos make up the larger French Polynesia archipelago:
Gambier, Tuamotu, Marquesas, Society, and Austral. Raivavae and Tubuai are located
within the southwestern extent of the Austral Islands archipelago, located between 23°18’
- 23°54’ S, and 149°34’ – 147°34’ W (Fig. 1A, B, and C). Bora Bora is located in the
western part of the Society archipelago between 16°26’ - 16°34’ S, and 151°47’ –
151°42’ W (Fig. 1A, D). The French Polynesia archipelago is located within the
southeastern trade wind belt and experiences prevailing winds from the east southeast;
however, swell is predominantly from the south southwest (Wisuki, 2012b, 2012a).
All three platforms possess a central, remnant, volcanic island and are surrounded
by annular reef rims which form near-continuous barriers around their lagoons (Fig. 1B,
C, and D).Conspicuously, the most prominent passes in the rim of both Raivavae and
Tubuai are through their northern margins. Raivavae has a secondary pass in the southern
rim and Tubuai has two secondary passes in the southwestern margin. Bora Bora possess
a singular pass located on the western margin of the platform. Both Raivavae and Tubuai
have several sand-cays, locally known as motus, atop their eastern, windward margins,
while Bora Bora possess motus along its eastern, windward margin, and northern margin.
Lagoonward of the reef crest, a back reef apron (maximum dip extent: Raivavae ~1.64
km; Tubuai ~2.32 km and Bora Bora ~ 2.64 km) grades gently into a deep lagoon
(maximum depth: Raivavae and Tubuai ~ 20 m and Bora Bora ~ 40 m (Gischler, 2011))
as shown in Figures 2A, B, and 3.
Figure 1: Location of French Polynesia in the central tropical Pacific Ocean (A). Inset
shows the location of Raivavae, Tubuai, and Bora Bora in relation to the French
Polynesian archipelago. WorldView-2 satellite imagery of Raivavae (B) and Tubuai (C),
and SPOT satellite imagery of Bora Bora (ESRI) (D) show each platform to possess a
fully aggraded reef margin surrounding a deep lagoon occupied by a remnant volcanic
island.
Figure 2: Bathymetric maps of Raivavae (A) and Tubuai (B) derived from spectral
analysis of WorldView-2 satellite imagery calibrated by single-beam acoustic depth
soundings acquired in the field.
Figure 3: SPOT satellite imagery (ESRI) with overlain sediment sample locations that are
color coded according to depth. Sample location and water depth obtained from Gischler
(2011).
3. Methods
3.1 Sediment Sample Collection
Surficial sediment samples were collected from Raivavae (n = 28) and Tubuai (n
= 21) as part of the Khaled Bin Sultan Living Oceans Foundation Global Reef Expedition
in April 2013 (Fig. 4). Sediment samples were collected with a handcrafted sediment
sampler (Fig. 5) made of a hollow metal cylinder with a fine meshed filter (< 0.03 mm)
attached to the end. The sediment sampler was attached to a line, deployed from the deck
of the boat, and dragged a few meters to ensure an adequate amount of sample was
collected. As the sampler was pulled through the sediment, water flowed through the
front opening of the sampler and out of the rear opening. Any sediment larger than 0.03
mm was retained in the sampler. Once aboard the boat, each sample was carefully
transferred from the sampler to a 100 ml Nalgene bottle. GPS coordinates were recorded
for each sample location. Water depth at the sample location was recorded using a single
bean acoustic depth sounder (SyQwest, Inc.). At the end of each sampling day, the
collected samples were decanted and dried in the main research vessel’s laboratory oven
at 70°C for a 24 hour period. Fully dried samples were necessary for granularmetric and
petrographic analysis of the sediment samples.
Figure 4: Worldview 2 satellite imagery of Raivavae (A), Tubuai (B) with overlain
sample locations.
Figure 5: Photographs of the hand crafted sediment sampler (A and B) including an up-
close view of the inside of the sampler showing the internal mesh filter (C).
3.2 Granularmetric Analysis of the Sediment Samples
Granularmetric analysis of the collected sediment samples was preformed to
ascertain the grain size (s) distribution and sorting. For the purposes of this study, grain
size distribution was described via percentages of gravel (s > 2mm), sand (0.062 mm < s
< 2mm), and mud (s < 0.063 mm). Each sample was weighed as a whole, emptied into a
stack of two sieves with mesh sizes of 2 mm and 0.063 mm, respectively, and placed on a
sieve shaker for five minutes to partition the sample into gravel, sand, and mud sized
fractions. Each fraction was then weighed separately and calculated as a percentage of the
whole sample. Next, sediment samples were analyzed for sorting using data measured by
a CAMSIZER (Retsch Technology, Haan, Germany). The CAMSIZER is a particle size
analyzer that utilizes a dual camera system to capture particle sizes ranging from 0.030
mm to 30 mm (Fig. 6). The CAMSIZER had limited capabilities to measure fine grains (s
< 0.063 mm), so only the gravel and sand (coarse) size fractions were processed through
the CAMSIZER. Data collected from the CAMSIZER were imported into GRADISTAT
v8 (Blott & Pye, 2001) to calculate sorting (Folk & Ward, 1957) of the coarse size
fraction of each sample (Table 1) .
Figure 6: CAMSIZER by Retsch Technology. Sediment sample is poured into the sample
funnel (A) that funnels the sediment onto the sample feeder (B). Vibrations from the
sample feeder transport grains across the feeder and over the feeder’s edge where they
cascade into the measurement shaft (C). The illumination unit (D) lights up the
measurement shaft, and allows real time recording of grains as they fall into the
measurement shaft. The basic camera (E) captures grains 0.300 mm – 30 mm in diameter,
while the zoom camera (F) captures grains 0.030 mm – 3 mm in diameter. Picture frames
(G) and (I) illustrate measurements from the basic and zoom camera respectively, while
picture frame (H) illustrates their combined measurements. (Image courtesy of
http://www,retsch-technology.com).
Table 1: Sorting classifications and their corresponding phi values based on Folk & Ward
(1957). Phi is the negative log of the diameter (mm) of a sediment grain.
Sorting Classification Range in Phi (φ)
Very well sorted 0.00 - 0.35
Well sorted 0.35 - 0.50
Moderately well sorted 0.50 - 0.71
Moderately sorted 0.71 - 1.00
Poorly sorted 1.00 - 2.00
Very poorly sorted 2.00 - 4.00
3.3 Petrographic Analysis of the Sediment Samples
Petrographic analysis was preformed to ascertain the faunal grain type
composition of each collected sediment sample. The sand fraction of each sample was
separated from the gravel with a 2 mm sieve. Only the sand-size fraction was used for
petrographic analysis due to the ubiquitously low abundance of gravel and mud. After
separation, a sediment splitter split the sand fraction of each sample four times to obtain a
6.25 ml sub-sample of the original 100 ml sample (Fig. 7). Sub-samples containing
greater than 50% sand sized grains less than 0.250 mm (very fine sand) were too fine to
be analyzed under a binocular microscope. These samples (n = 15) were sent to National
Petrographic Service, Inc. to be made into thin sections that could be examined and point-
counted with a petrographic microscope. All other samples were point counted as loose
grains under a binocular microscope.
Figure 7: Illustration of the steps used to split a sample in preparation for point-counting
loose sediment for faunal grain types. The sample is split four times to create a subsample
that is 16 times smaller than the original sample (100 ml to 6.25 ml). The sample is then
spread out uniformly on the counting tray to be examined under a binocular microscope.
Point-counting was used to calculate the faunal grain type composition of each
sediment sample utilizing 11 biogenic categories: bryozoan, coral, coralline algae,
mollusk, Halimeda, echinoderm, Foraminifera (foram), octocoral spicule (spicule),
serpulid, crustacean and unknown (Appendix A). The thin section samples (n = 15) were
photographed with a microscope slide scanner PathScan Enabler IV (Electron
Microscopy Sciences, Hatfield, PA) to create a digital image of the slide (Fig. 8). These
images were imported into Coral Point Count with Excel extensions (CPCe) (Kohler &
Gill, 2006) to create a uniform 400 point reference grid over each petrographic scan. A
petrographic microscope was used to identify grain types in thin section. This was done
by cross-referencing each point on the 400 point digital reference grid with the
corresponding point on the thin section. This was repeated until 200 grains were
identified. Blank spaces were skipped. The remaining samples (n = 34), were uniformly
spread onto a small rectangular tray with a transparent 200 point grid overlain atop the
sample (Fig. 7). A binocular microscope was used to visually identify grains at each grid
point for each sample. The abundance of each faunal grain type was recorded as a relative
percentage of the sample. Literature by Scholle and Ulmer-Scholle (2003) and Flügel
(2004) were consulted when identifying grain types.
Figure 8: Photomicrograph of a thin section (sample FPA-16) taken by PathScan Enabler
IV (A). Grains are black, grey, brown, and white. The blue background is the epoxy used
to create the thin section. The large holes in the thin section are air bubbles that formed
when the epoxy cured. The red circles highlight the location of examples (B) – (E).
Halimeda fragment (B), characterized by reddish brown color and porous internal
structure. Mollusk shell (C), characterized by large partitioned chambers. Coral fragment
(D), characterized by a white skeleton and brown skeletal interstitial space filled with
mud-sized sediments. Coralline algae fragment (E), characterized by reddish brown color
and concentric cells that radiate from the center of the grain.
3.4 Delineating the Platform Margin and Platform Interior from Satellite
Imagery
Satellite imagery was interpreted to manually delineate the platform margin
(margin), the platform interior (interior), and the central island of Raivavae, Tubuai, and
Bora Bora using ArcMap10.3 (ESRI). Sediment samples were assigned to either the
margin or interior based on these delineations. For the purpose of this thesis, the margin
was defined as the zone extending from the reef rim to the lagoonal termination of the
back reef sand apron (LTBRA) (Fig. 9B, C). The reef rim was defined from the satellite
imagery as the transition from the dark brown reef flat and the open ocean, and
characterized by breaking waves (Fig. 9A, B). The LTBRA was defined as the sharp
color change between the light-blue back reef apron and dark-blue lagoon (Fig. 9A, B).
The interior was delineated as the lagoon, spanning from the LTBRA to the central island
(Fig. 9B, C). The central island was delineated by the perimeter of its shoreline (Fig. 9B).
Figure 9: WorldView-2 image subscenes of Raivavae showing an example of the imagery
used to delineate the reef rim, LTBRA, and the central island (A); delineations (yellow)
of the reef rim, the LTBRA, and the central island (B); and extents of the margin and
interior as confined by these delineations (C).
3.5 Calculating Relative Distance of a Sediment Sample from the Reef Rim
Relative distance was used as a measure to quantify the distance of a sediment
sample from the reef rim. Using relative distance allowed for easier comparison of
distance between the three different sized platforms. First, to calculate relative distance
from the reef rim (RDRR), the shortest straight line distance from the delineation of the
reef rim to a sediment sample location was measured as a transect in GIS (Fig. 10A).
Second, the transect was extended past the sediment sample location to the delineation of
the central island to measure the total length of the transect (Fig. 10B). Finally, these two
measurements were used in the following equation to calculate RDRR (Fig. 10C):
RDRR = D1
D2 (1)
where D1 is the distance measured from the delineation of the reef rim to a sediment
sample location, and D2 is the total distance of the respective transect spanning from the
reef rim to the central island.
Figure 10: WorldView-2 Image subscenes of Raivavae demonstrating the process of
measuring the distance from the reef rim to a sediment sample location, D1, (A),
measuring the distance of the transect spanning from the reef rim to central island, D2,
(B) and calculating RDRR from D1 and D2 (C). The white lines are transects with arrows
indicating the direction in which each transect spans. The yellow circles represent
sediment sample locations.
4 Statistical Methods
4.1 Formulating Models to Predict Water Depth and RDRR
A set of 22 statistical models were formulated to test how well the response
variables, water depth and RDRR, could be predicted based on sediment character. Six
sedimentary properties were used as explanatory variables: percent gravel, percent sand,
percent mud, sorting, percent coral, and percent Halimeda. Of the eleven faunal grain
type categories, only coral and Halimeda were used in the statistical modeling because
they were present in all sediment samples, and these fauna can be associated with
particular environments of deposition. In particular, coral is prevalent in marginal, reef
rim environments while Halimeda has been reported as a common constituent within
interior, lagoonal environments (Hillis-Colinvaux, 1980; Braga et al., 1996; Chevillon,
1996; Yamano et al., 2002; Montaggioni, 2005; Tucker & Wright, 2009; Rankey et al.,
2011; Wasserman & Rankey, 2014). However, Halimeda has also been reported to be a
major constituent in marginal settings of some reef systems. Consequently, the
abundance of Halimeda within the margin and interior varies between reef systems
(Gischler, 2011).
For each of the two response variables, the set of 22 statistical models included
six linear models, 15 multilinear models, and one random model. The set of models was
limited to two sedimentary properties because the low sample size of the study sites
(Raivavae: n = 28, Tubuai: n = 21, and Bora Bora: n = 31) would result in statistically
unreliable models for three or more explanatory variables (Babyak, 2004). The linear
models took the form of:
y = β0+ β1x1 (2)
where y is the response variable (water depth or RDRR), x1 is the explanatory variable
(one of the six sedimentary properties), and the variables β0 and β1 are statistical
coefficients fitted through linear regression. The multilinear models are similarly
structured such that two explanatory variables are paired. They take the form of:
y = β0+ β1x1+ β2x2 (3)
where x2 is a second explanatory variable and β2 is a statistical coefficient fitted through
multilinear regression. Finally, the random model takes the form:
y = β0 (4)
such that the response variable y is not related to an explanatory variable.
The two sets of 22 models were used for the Raivavae and Tubuai datasets. For
Bora Bora, two sets of only four models were used, because percent coral and percent
Halimeda were the only two sedimentary properties from the Gischler (2011) dataset that
were readily comparable to the Raivavae and Tubuai datasets. Three linear models:
percent coral, percent Halimeda, and random, and one multi-linear model: percent coral
and percent Halimeda were used for Bora Bora.
4.2 Applying Models to the Study Sites
The two sets of models were applied to each study site dataset. First, each dataset
was partitioned into three subgroups: platform-wide (i.e. all samples), margin (i.e. margin
samples), and interior (i.e. interior samples). This allowed the models to be assessed on
their ability to predict water depth and RDRR for the entire platform, the margin, and the
interior separately. The lm() function (i.e. regression analysis) in R 3.0 was used to
estimate a line of best fit for each model as applied to each dataset subgroup (Team,
2014). The line of best fit was estimated based on the set of explanatory and observed
response variables for each dataset subgroup. The statistical coefficients and a new set of
predicted response variables for each model were estimated from the line of best fit.
4.3 Assessing the Accuracy of the Predictive Models
The accuracy of each model was evaluated by calculating the root mean squared
error (RMSE). The RMSE gives a mean value of the error, or difference, between the
predicted and observed response variables (Burnham & Anderson, 2002). Low RMSE
values indicate little difference between the predicted and observed variables, and thus
indicate a more accurate model. RMSE was calculated with the following equation:
RMSE = √∑ (ŷi - yi)
2ni=1
n (5)
where ŷi is the set of predicted response variables, yi is the set of observed response
variables, and n is the sample size. RMSE for models predicting RDRR was on a scale of
0 – 1 (0% - 100% of total transect length), while RMSE for models predicting water
depth was in meters. The RMSE of each model was then normalized (NRMSE) to the
range of water depth or RDRR measurements within the respective subgroup.
Normalizing the RMSE provided a context for the accuracy of each model given these
ranges. NRMSE was on a scale of 0 – 1, and gives the overall error of the model.
NRMSE = RMSE
ymax - ymin (6)
Based on NRMSE, each set of models for each dataset subgroup of Raivavae and
Tubuai were narrowed down to the five most accurate models. Out of these five models,
the model that showed the greatest accuracy across all three subgroups was selected as
the most applicable model for the study site. Bora Bora only had four models to choose
from, so the one model out of the four that showed the greatest accuracy across all three
subgroups was selected.
4.4 Applying Raivavae Models to Tubuai and Bora Bora
Raivavae models were also tested on Tubuai and Bora Bora to examine their
aptitude to predict water depth and RDRR on other, similar, isolated carbonate platforms.
The one-water depth and RDRR model selected from the top five Raivavae models were
tested on Tubuai. For Bora Bora, the most accurate Raivavae water depth and RDRR
model out of the set of four applicable models: percent coral, percent Halimeda, random,
and percent coral and percent Halimeda was applied. The model coefficients from the
Raivavae models were kept, while the explanatory variables from Tubuai and Bora Bora
were used. Running the models with Raivavae coefficients and Tubuai and Bora Bora
explanatory variables allowed for a new set of response variables to be estimated for each
site. RMSE was applied using the new set of response variables, and NRMSE calculated
from RMSE. The accuracy of the Raivavae models as applied to Tubuai and Bora Bora
would thus reveal if statistical trends observed from Raivavae were also apparent for
Tubuai and Bora Bora.
4.5 Linear Discriminant Analysis
Linear discriminant analysis (LDA) was used to test how accurately a
sedimentary property or combination of sedimentary properties could differentiate
between the margin and interior environments. LDA is a statistical means of using a
linear combination of variables to separate two or more classes of data (Fisher, 1936).
This metric is similar to the ANOVA test in that it attempts to define a response variable
based on a linear combination of explanatory variables (Fisher, 1936; McLachlan, 2004).
However, LDA differs from ANOVA in that it uses continuous data (e.g. percent sand) as
the explanatory variables and categorical data (e.g. margin or interior) as the response
variables; whereas the methodology of ANOVA is the opposite (Wetcher-Hendricks,
2011). First, box and whisker plots of the abundance of each sedimentary property by
each subgroup were analyzed to gauge which of the six candidate sedimentary properties
would likely show differentiation by the margin and interior. Any sedimentary properties
that showed no overlap of the lower and of the upper quartiles (i.e. upper and lower
whiskers) when comparing between margin and interior were chosen to be used as a
variable in LDA because they had the least overlap of data and would most likely
produce the best differentiation between the margin and interior. The selected
sedimentary properties were then subjected to LDA to formulate a model to differentiate
between the margin and interior.
Leave one out cross validation (LOOCV) was used to test the accuracy of the
formulated LDA model. LOOCV works by removing one entry from the dataset and
training the model to best predict the removed entry (Lachenbruch & Mickey, 1968).
This procedure is repeated for each entry in the dataset to estimate the overall accuracy of
the model. This methodology was used to formulate a LDA model for each of three
datasets. The LDA model from each dataset was additionally tested on the other two
datasets to evaluate if a singular LDA model could be used among platforms.
5. Results
5.1 Sedimentary Properties
5.1.1 Raivavae and Tubuai
The textural character of Raivavae and Tubuai sediments were similar in many
ways; however, there were clear differences that separate sediments from these two
platforms. Raivavae and Tubuai can be characterized by a prominence of sand sized
grains, both in the margin and interior. Gravel was far less abundant, while mud was
scarce, on both (Fig. 11A, B and Appendix B, Tables 1 and 3). On Raivavae, sand
decreased in abundance from the margin (mean = 85.59%) to the interior (mean =
79.86%), while the abundance of sand remained consistent from the margin (mean =
76.07%) to the interior (mean = 78.33%) of Tubuai. The decrease in sand in the interior
of Raivavae was balanced by an increase in mud (mean: margin = 0.36% and interior =
7.34%). Gravel remained virtually constant between the margin (mean = 14.14%) and
interior (mean = 12.71%) of Raivavae. In contrast, gravel decreased in abundance from
the margin (mean = 23.53%) to the interior (12.00%) of Tubuai. This decrease was
balanced with an increase in mud in Tubuai’s interior (mean: margin = 0.40% and
interior = 9.67%). The increase in the abundance of mud in the interior of Tubuai was
primarily influenced by high concentrations of mud measured from sediment samples
FPA-57 (mud = 18.00%) and FPA-70 (mud = 35.00%). Sorting of Raivavae sediments
range from poorly sorted to very poorly sorted, with margin and interior sediments
averaging as very poorly sorted (Fig. 12A and Appendix B, Table 1). Sorting of Tubuai
sediments ranged from moderately sorted to poorly sorted, with margin and interior
sediments averaging as poorly sorted (Fig. 12B and Appendix B, Table 3).
Faunal grain types observed in Raivavae and Tubuai sediments were very similar
as well; however, there were also clear differences that distinguished these two sites (Fig.
13A and B and Appendix B, Tables 2 and 4). The main faunal grain types observed from
both sites were: coral (mean: Raivavae = 27.21% and Tubuai = 23.55%), coralline algae
(mean: Raivavae = 9.96% and Tubuai = 11.35%), Halimeda (mean: Raivavae = 33.61%
and Tubuai = 27.20%), and mollusk (mean: Raivavae = 19.43% and Tubuai = 26.05%).
For both sites the abundance of coral and mollusks were the greatest in the margin. On
Raivavae, coral showed a marked decrease in abundance from the margin (mean =
34.57%) to the interior (mean = 19.86%), while mollusks showed a lesser decrease
(mean: margin = 21.57% and interior = 17.29%). On Tubuai, the decrease in coral from
the margin to the interior was not a drastic (margin mean = 24.60% and interior mean =
20.83%). The same was true for the abundance of mollusks (mean: margin = 27.00% and
interior = 23.33%). Coralline algae showed a greater abundance in Raivavae’s margin
(mean = 11.07%) than the interior (mean = 8.86%). In contrast, coralline algae displayed
a slight increase in abundance from the margin (11.07%) to the interior (12.33%) of
Tubuai. For both sites, Halimeda was greatest in the interior (mean: Raivavae = 43.79%
and Tubuai = 33.00%) with a decrease in the margin (mean: Raivavae = 23.43% and
Tubuai = 24.80%). The other faunal grain types (bryozoan, echinoderm, foram, spicule,
serpulid, and crustacean) observed in Raivavae and Tubuai sediment samples did not
exceed a mean of more than 2.36% and 3.87%, respectively. Unknown grains had a mean
of 6.50% and 7.15% for Raivavae and Tubuai, respectively.
5.1.2 Bora Bora
The only sedimentary data obtained for Bora Bora was the abundance of coral and
Halimeda. These data, along with water depth recordings for the Bora Bora sediment
samples, were obtained from Gischler (2011). Appendix B, Table 5 contains this data
along with measurements of RDRR for each Bora Bora sediment sample.
Figure 11: WorldView-2 satellite imagery of Raivavae (B) and Tubuai (B) with overlain
bar charts that represent the abundance of gravel (red), sand (yellow) and mud (green) of
every collected sediment sample.
Figure 12: WorldView-2 satellite imagery of Raivavae (A) and Tubuai (B) with overlain
circles that represent the sorting of every collected sediment sample.
Figure 13: WorldView-2 satellite imagery of Raivavae (A) and Tubuai (B) with overlain
bar charts that represent the faunal grain type composition of every collected sediment
sample.
5.2 Sedimentary Properties Showed Mixed Performance for Predicting Water
Depth
5.2.1 Raivavae and Tubuai Models
Overall, the selected sedimentary properties showed the greatest applicability to
predict water depth within the platform-wide and margin subgroups of Raivavae and the
platform-wide and interior subgroups of Tubuai (Appendix B, Tables 6 and 7). The
model that could be applied most accurately to all three subgroups of the Raivavae
dataset predicted water depth based on percent mud and percent coral (NRMSE:
platform-wide = 0.19, margin = 0.19, and interior = 0.23), as seen in Table 2. Applying
this model to the Tubuai dataset resulted less accurate predictions (NRMSE: platform-
wide = 0.22, margin = 0.31, and interior = 0.37). Conversely, the Tubuai model
performed with greater accuracy (NRMSE: platform-wide = 0.11, margin = 0.21, and
interior = 0.07), as seen in Table 3. This model predicted water depth based on percent
gravel and percent mud.
Table 2: Summary of the top selected Raivavae water depth model as applied to the
platform-wide, margin, and interior subgroups of the Raivavae and Tubuai datasets.
Coefficients of the model vary by subgroup. RMSE of the model is given in meters.
Depth range is also in meters, and represents the depth range of sediment samples used in
the model. NRMSE (in bold) gives the error (on a scale of 0 – 1) of the model for the
given depth range.
Top Selected Raivavae Water Depth Model as Applied to Raivavae and Tubuai
Subgroup Platform-wide Margin Interior
Model DEPTH = 8.130 +
9.779MU – 15.243CR
DEPTH = - 0.022 +
16.079MU +
5.093CR
DEPTH = 0.563 +
0.735MU –
0.012CR
As Applied to
Raivavae
RMSE (m) 2.69 0.69 2.65
Depth
Range (m) 0.72 - 14.88 0.72 - 2.56 3.52 - 14.88
NRMSE 0.19 0.19 0.23
As Applied to
Tubuai
RMSE (m) 4.79 3.53 7.47
Depth
Range (m) 0.74 - 21.76 0.74 - 12.11 1.64 - 21.76
NRMSE 0.22 0.31 0.37
Note. The abbreviations MU and CR represent percent mud and percent coral, respectively
Table 3: Summary of the top selected Tubuai water depth model as applied to the
platform-wide, margin, and interior subgroups of the Tubuai dataset. Coefficients of the
model vary by subgroup. RMSE of the model is given in meters. Depth range is also in
meters, and represents the depth range of sediment samples used in the model. NRMSE
(in bold) gives the error (on a scale of 0 – 1) of the model for the given depth range.
Top Selected Tubuai Water Depth Model as Applied to Tubuai
Subgroup Platform-wide Margin Interior
Model Depth = 6.033 - 11.280GR
+ 48.765MU
Depth = 5.984 – 10.461GR
– 195.764MU
Depth = 7.034 –
17.485GR + 46.613MU
RMSE (m) 2.27 2.43 1.48
Depth Range
(m) 0.74 - 21.76 0.74 - 12.11 1.64 - 21.76
NRMSE 0.11 0.21 0.07
Note. The abbreviations GR and MU represent percent gravel and percent mud, respectively
5.2.2 Raivavae and Bora Bora Models
Overall, coral and Halimeda were not accurate predictors of water depth for the
Bora Bora dataset (Appendix B, Table 8). The most accurate model formulated from the
Raivavae dataset suitable to be tested on the Bora Bora dataset predicted water depth
based on percent coral and percent Halimeda (Table 4). This model showed variable
accuracy as applied to the Raivavae subgroups (NRMSE: platform-wide = 0.19, margin =
0.38, and interior = 0.23). The Raivavae model showed a decrease in performance as
applied to Bora Bora (NRMS: platform-wide = 0.44, margin = 0.31, and interior = 0.58).
Comparatively, the Bora Bora model produced more accurate predictions of water depth
(NRMSE: platform-wide = 0.25, margin = 0.27, and interior = 0.18), as seen in Table 5.
This model also predicted water depth based on percent coral and percent Halimeda, but
had different coefficients than the Raivavae model.
Table 4: Summary of the top selected Raivavae water depth model suitable to be applied
to the Raivavae and Bora Bora datasets. The model was applied to the platform-wide,
margin, and interior subgroups of each dataset. Coefficients of the model vary by
subgroup. RMSE of the model is given in meters. Depth range is also in meters, and
represents the depth range of sediment samples used in the model. NRMSE (in bold)
gives the error (on a scale of 0 – 1) of the model for the given depth range.
Top Selected Raivavae Water Depth Model Suitable to be Applied to Raivavae and Bora Bora
Subgroup Platform-wide Margin Interior
Model Depth = 6.629 -
12.415CR + 3.278HA
Depth = 1.146 +
3.398CR - 2.387HA
Depth = 10.890 -
14.725CR -
2.337HA
As Applied to
Raivavae
RMSE (m) 2.72 0.7 2.68
Depth
Range (m) 0.72 - 14.88 0.72 - 2.56 3.52 - 14.88
NRMSE 0.19 0.38 0.24
As Applied to
Bora Bora
RMSE (m) 16.41 7.85 16.35
Depth
Range (m) 0 - 37 0 - 25 0 - 37
RRMSE 0.44 0.31 0.58
Note. The abbreviations CR and HA represent percent coral and percent Halimeda, respectively
Table 5: Summary of the top selected Bora Bora water depth model as applied to the
platform-wide, margin, and interior subgroups of the Bora Bora dataset. Coefficients of
the model vary by subgroup. RMSE of the model is given in meters. Depth range is also
in meters, and represents the depth range of sediment samples used in the model.
NRMSE (in bold) gives the error (on a scale of 0 – 1) of the model for the given depth
range.
Top Selected Bora Bora Water Depth Model as Applied to Bora Bora
Subgroup Platform-wide Margin Interior
Model Depth = 25.187 -
31.596CR - 28.822HA
Depth = 6.793 - 4.257CR -
3.912HA
Depth = 28.026 -
34.680CR + 3.535HA
RMSE (m) 9.14 6.86 4.92
Depth Range
(m) 0 - 37 0 - 25 0 - 37
NRMSE 0.25 0.27 0.18
Note. The abbreviations CR and HA represent percent coral and percent Halimeda, respectively
5.3 Sedimentary Properties Show Moderate Accuracy for Predicting Relative
Distance from the Reef Rim
5.3.1 Raivavae and Tubuai Models
The selected sedimentary properties showed an overall decreased performance for
predicting RDRR than for water depth for Raivavae and Tubuai (Appendix, Tables 9 and
10). Also, much like the water depth models, the most accurate predictions came from
site-specific models. The model that could be applied most accurately to all three
subgroups of the Raivavae dataset predicted RDRR based on percent sand and percent
coral (Table 6). This model showed moderate accuracy for all three subgroups (NRMSE:
platform-wide: 0.23, margin = 0.27, and interior = 0.24). The Raivavae model performed
less accurately on Tubuai (NRMSE: platform-wide = 0.35, margin = 0.42, and interior =
0.39). Compared to the Raivavae model, the Tubuai model showed an increased accuracy
(NRMSE: platform-wide = 0.32, margin = 0.33, and interior = 0.20), as seen in Table 7.
This model predicted RDRR based on percent gravel and sorting.
Table 6: Summary of the top selected Raivavae RDRR model as applied to the platform-
wide, margin, and interior subgroups of the Raivavae and Tubuai datasets. Coefficients of
the model vary by subgroup. RMSE of the model is on a scale of 0 – 1, and represents
0% – 100% of total transect length spanning from the reef rim to the central island.
NRMSE (in bold) gives the error (on a scale of 0 – 1) of the model for the range of
RDRR measurements.
Top Selected Raivavae RDRR Model as Applied to Raivavae and Tubuai
Subgroup Platform-wide Margin Interior
Model RDRR = 0.460 +
0.397SA – 0.993CR
RDRR = - 0.480 +
1.263SA - 0.468CR
RDRR = 0.900 –
0.098SA –
1.171CR
As Applied to
Raivavae
RMSE 0.17 0.15 0.15
Range of
RDRR
Measurements
0.17 – 0.89 0.17 – 0.72 0.27 – 0.89
NRMSE 0.23 0.27 0.24
As Applied to
Tubuai
RMSE 0.27 0.33 0.13
Range of
RDRR
Measurements
0.09 – 0.86 0.09 – 0.86 0.45 – 0.81
NRMSE 0.35 0.43 0.39
Note. The abbreviations SA and CR represent percent sand and percent coral, respectively
Table 7: Summary of the top selected Tubuai RDRR model as applied to the platform-
wide, margin, and interior subgroups of the Tubuai dataset. Coefficients of the model
vary by subgroup. RMSE of the model is on a scale of 0 – 1, and represents 0% – 100%
of total transect length spanning from the reef rim to the central island. NRMSE (in bold)
gives the error (on a scale of 0 – 1) of the model for the range of RDRR measurements.
Top Selected Tubuai RDRR Model as Applied to Tubuai
Subgroup Platform-wide Margin Interior
Model RDRR = 0.163 –
0.373GR + 0.352SO
RDRR = 0.131 –
0.334GR + 0.353SO
RDRR = 0.562 –
0.735GR +
0.012SO
RMSE 0.25 0.26 0.06
Range of RDRR Measurements 0.09 – 0.86 0.09 – 0.86 0.45 – 0.81
NRMSE 0.32 0.33 0.20
Note. The abbreviations GR and SO represent percent gravel and sorting, respectively
5.3.2 Raivavae and Bora Bora Models
Coral and Halimeda showed similar performance for predicting RDRR for Bora
Bora as they did for water depth; with an increase in performance within the interior
subgroup (Appendix, Table 11). The most accurate model formulated from the Raivavae
dataset suitable to be tested on the Bora Bora dataset predicted RDRR based on percent
coral and percent Halimeda (Table 8). This model had moderate accuracy as applied to
Raivavae (NRMSE: platform-wide = 0.24, margin = 0.32, and interior = 0.24). Applying
this to Bora Bora yielded equal results for the platform-wide subgroup (NRMSE = 0.24),
but less accurate results for the margin and interior subgroups (NRMSE: margin 0.55=
and interior = 0.32). Compared to the Raivavae model, the Bora Bora produced equally
accurate predictions for the platform-wide subgroup (NRMSE = 0.24), more accurate
predictions for the margin subgroup (NRMSE = 0.25), and less accurate predictions for
the interior subgroup (NRMSE = 0.26), as seen in Table 9. This model also predicted
RDRR based on percent coral and percent Halimeda, but had different coefficients than
the Raivavae model.
Table 8: Summary of the top selected Raivavae RDRR model suitable to be applied to the
Raivavae and Bora Bora datasets. The model was applied to the platform-wide,
MARGIN, and INTERIOR subgroups of each dataset. Coefficients of the model vary by
subgroup. RMSE of the model is on a scale of 0 – 1, and represents 0% – 100% of total
transect length spanning from the reef rim to the central island. NRMSE (in bold) gives
the error (on a scale of 0 – 1) of the model for the range of RDRR measurements.
Top Selected Raivavae RDRR Model Suitable to be Applied to Raivavae and Bora Bora
Subgroup Platform-wide Margin Interior
Model
RDRR = 0.70 -
0.801CR +
0.097HA
RDRR = 1.019 -
1.033CR - 0.953HA
RDRR = 0.645 -
0.809CR +
0.241HA
As Applied to
Raivavae
RMSE 0.17 0.18 0.15
Range of
RDRR
Measurements
0.17 – 0.89 0.17 – 0.72 0.27 – 0.89
NRMSE 0.24 0.32 0.24
As Applied to
Bora Bora
RMSE 0.21 0.36 0.22
Range of
RDRR
Measurements
0.07 – 0.96 0.07 – 0.73 0.28 – 0.96
NRMSE 0.24 0.55 0.32
Note. The abbreviations CR and HA represent percent coral and percent Halimeda, respectively
Table 9: Summary of the top selected Bora Bora RDRR model as applied to the platform-
wide, margin, and interior subgroups of the Bora Bora dataset. Coefficients of the model
vary by subgroup. RMSE of the model is on a scale of 0 – 1, and represents 0% – 100%
of total transect length spanning from the reef rim to the central island NRMSE (in bold)
gives the error (on a scale of 0 – 1) of the model for the range of RDRR measurements.
Top Selected Bora Bora RDRR Model Applied to Bora Bora
Subgroup Platform-wide Margin Interior
Model RDRR = 0.678 –
0.621CR – 0.081HA
RDRR = 0.453 –
0.853CR – 0.758HA
RDRR = 0.697 –
0.316CR –
0.328HA
RMSE 0.21 0.17 0.18
Range of RDRR
Measurements 0.07 – 0.96 0.07 – 0.73 0.28 – 0.96
NRMSE 0.24 0.25 0.26
Note. The abbreviations CR and HA represent percent coral and percent Halimeda, respectively
5.4 Sedimentary Properties Can’t Reliably Differentiate Between Platform
Margin and Platform Interior
5.4.1 Raivavae Linear Discriminant Analysis Model
Box plots for each of the six sedimentary properties were analyzed, and revealed
coral and Halimeda to have the greatest variance in abundance when comparing between
the margin and interior (Fig. 14). Thus, these properties were selected for further analysis
using LDA. The resulting LDA model formulated from the Raivavae dataset was:
R = 7.924CR – 1.920HA (7)
where R is the Raivavae LDA model and CR and HA are the percentage of coral and
Halimeda, respectively. This model was used as a mathematical boundary line to classify
sediment samples as either belonging to the margin or interior environment of deposition
based on the abundance of coral and Halimeda measured from the sample. The model
had an accuracy of 78.57% and 71.43% for the margin and interior of Raivavae,
respectively (Fig. 15A). It produced stronger results for the interior samples of Tubuai
(accuracy = 83.33%) and Bora Bora (accuracy = 84.21%), but weaker results for the
margin samples (accuracy: Tubuai = 46.67% and Bora Bora = 45.45%) (Fig. 15B, C).
Figure 14: Box plots illustrating the variance in abundance of each sedimentary property
between the margin (M) and interior (I) of Raivavae (A), Tubuai (B), and Bora Bora (C).
Coral and Halimeda showed the greatest variance in abundance between the margin and
interior for all three study sites, as illustrated by the degree of separation between the
margin and interior box and whiskers for each grain type.
Margin Interior
Margin 11 4
Interior 3 10
Accuracy 78.57% 71.43%
LDA
Margin Interior
Margin 7 1
Interior 8 5
Accuracy 46.67% 83.33%
LDA
Figure 15: Representations of the margin (green) and interior (red) drawn over satellite
imagery coupled with boxplots that show the abundance of coral and Halimeda for the
margin (green) and interior (red) of Raivavae (A), Tubuai (B), and Bora Bora (C).
Results from applying the LDA model formulated from the Raivavae dataset to each of
the three platforms are included in the associated tables. Correct classifications are in
bold.
5.4.2 Tubuai and Bora Bora Linear Discriminant Analysis Models
Two other LDA models were formulated using the Tubuai dataset and the Bora
Bora dataset. The Tubuai LDA model was:
T = -1.766CR - 10.394HA (8)
where T is the Tubuai LDA model and CR and HA are the percentage of coral and
Halimeda, respectively. The Bora Bora LDA model was:
B = 1.892CR + 8.688HA (9)
where B is the Bora Bora LDA model and CR and HA are the percentage of coral and
Halimeda, respectively. Again, these models were formulated to classify sediment
Margin Interior
Margin 5 3
Interior 6 16
Accuracy 45.45% 84.21%
LDA
samples as either belonging to the margin or interior environment of deposition based on
the abundance of coral and Halimeda measured from the sample.
The Tubuai LDA model produced strong results for classifying the margin
samples (accuracy: Raivavae = 92.86%, Tubuai = 93.33%, and Bora Bora = 90.91%), but
poor results for classifying the interior samples (accuracy: Raivavae = 57.14%, Tubuai =
16.67%, and Bora Bora = 0%), as shown in Table 10. Similarly, the Bora Bora specific
model produced strong results for the margin samples of Raivavae (accuracy = 92.85%)
and Tubuai (accuracy = 100%), and poor results for the interior samples of Raivavae and
Tubuai (accuracy: both = 0%) (Table 11). However, it produced stronger results for the
interior samples of Bora Bora (accuracy = 84.21%), while producing weaker results for
the margin samples of Bora Bora (accuracy = 54.55%). Of the three models that were
evaluated (equations 6 – 8), the Raivavae model (equation 6) produced the best overall
results for all three datasets (accuracy: Raivavae = 75%, Tubuai = 65%, and Bora Bora =
64.83%).
Table 10: Results from applying the Tubuai LDA model to all three datasets. Correct
classifications are in bold.
Raivavae Tubuai Bora Bora
Margin Interior Margin Interior Margin Interior
Margin 13 6 14 5 10 19
Interior 1 8 1 1 1 0
Accuracy 92.86% 57.14% 93.33% 16.67% 90.91% 0.00%
Table 11: Results from applying the Bora Bora LDA model to all three datasets. Correct
classifications are in bold.
Raivavae Tubuai Bora Bora
Margin Interior Margin Interior Margin Interior
Margin 13 14 15 6 6 3
Interior 1 0 0 0 5 16
Accuracy 92.86% 0.00% 100.00% 0.00% 54.55% 84.21%
7. Discussion
Results from this study indicate that the selected sedimentary properties can be
used to predict water depth and RDRR from the three study sites with moderate accuracy.
Generally, water depth was predicted with ≥ 73% accuracy, while RDRR was predicted
with ≥ 67% accuracy. Of course, the accuracy of water depth and RDRR predictions
varied among and within platforms. In addition, the models selected from each platform
as the most applicable were site specific. However, the Raivavae water depth model that
used mud and coral as explanatory variables was accurate for both Raivavae (accuracy =
81%) and Tubuai (accuracy = 78%). Percent coral and percent Halimeda exhibited
similar predictive power when used in the Raivavae LDA model to differentiate between
the margin and interior environments (accuracy: Raivavae: 75%, Tubuai = 65%, and Bora
Bora = 65%).
The level of accuracy presented by the water depth, RDRR, and LDA models is
not surprising. Though most isolated platforms display a general pattern of sediment
distribution (i.e. larger, coralgal grains proximal to the reef crest and finer sediments
constituted by a mixture of Halimeda, mollusk, and forams in the interior), much
variation in this pattern has been documented (Masse et al., 1989; Chevillon, 1996;
Gischler, 2006; Rankey & Reeder, 2010). The findings of this study present that, on these
three platforms, the distribution of sediments shows general trends with water depth and
distance from the reef rim. The predictive relationship between sedimentary properties
and water depth and distance from the rim is, however, most likely diminished by local
environmental effects that govern sediment production, redistribution, and accumulation
(e.g. assemblages of sediment producers, wind and wave energy, currents of removal and
currents of delivery, bioerosion and bioturbation, tidal velocity, and antecedent
topography). These effects vary in extent and duration across the platform top and work
together in unique ways to produce complex spatial patterns of sediment distribution
among and within isolated platforms (Wright & Burgess, 2005; Rankey et al., 2011;
Harris et al., 2014b).
7.1 Apparent Local Environmental Effects
Wave energy, in the form of local wind-driven waves, open ocean swell, and
storm surge is a well-documented environmental effect that influences sediment
distribution in reef environments (Rankey et al., 2011; Harris et al., 2014a; Wasserman &
Rankey, 2014; Purkis et al., 2015b). Wave energy interacts with each platform differently
to create among and within platform differences of sediment accumulation. Waves are
usually strongest along the windward margin of a platform (Hine et al., 1981); however,
distal, open ocean swell may propagate from a direction that is contrary to windward
influences and have a greater impact of sediment distribution (Rankey et al., 2009;
Wasserman & Rankey, 2014). Raivavae, Tubuai, and Bora Bora are all within the
southeast trade wind belt and experience prevailing winds from the east southeast
(Wisuki, 2012b, 2012a). In addition, all three of these platforms experience open ocean
swell from the south southwest (Wisuki, 2012b, 2012a). The influences from the trade
winds and open ocean swell can be observed in the large expanse of the eastern, southern,
and western sand aprons of each platform. Though each platform appears to have
generally similar windward and swell-ward influences, the distance of each isolated
paltform from one another (distance: Raivavae and Tubuai ~ 210 km, Raivavae and Bora
Bora ~ 920 km, and Tubuai and Bora Bora ~ 800 km) suggests that each site probably
experiences local variations of wind speed and direction and wave direction, height and
period that influences the accumulation of sediments in unique ways.
The manner in which local winds and waves interact with each platform and
influence sediment distribution is based on the morphology (i.e. size and shape) and
orientation of the platform in regards to prevailing winds and waves (Rankey & Garza-
Pérez, 2012). The three study sites exhibit a variety of size, shape, and orientation to
wind and wave influences. Raivavae is horizontally elongated, with the longest vertical
and horizontal axis having approximate dimensions of 7.36 km x 14.33 km. Raivavae is
oriented on a slight angle that completely exposes its southern margin to the east
southeast prevailing trade winds and associated wind driven waves. Influence from the
wind driven waves can be observed in the lagoonward extent of the southern sand apron.
The largest extent of which spans about 2 km from reef crest to lagoon. Conversely,
Tubuai and Bora Bora are both larger, less horizontally elongated (dimensions: Tubuai ~
10.73 km x 16.69 km and Bora Bora ~ 14.41 km x 9.87 km), and positioned on a north-
south orientation. Tubuai has a pronounced western sand apron (max extent ~ 2.5 km)
while Bora Bora has a pronounced southern sand apron (max extent ~ 2.5 km). The
lagoonward extents of both of these sand aprons are indicative of each platforms
exposure to southwestern open ocean swell.
The vertical and lateral extents of sediment accumulations produced by wave and
tidal energy can also be influenced by antecedent topography. The series of topographic
lows and highs of the antecedent topography act as sinks and barriers to sediment
accumulation, and as conduits and barriers to current flow (Rankey et al., 2009; Harris et
al., 2014a; Isaack & Gischler, 2015). The antecedent topography of a platform is a factor
of subaerial solution (karstification) of the bedrock subsurface during periods of sea-level
lowstands and differential accretion rates of platform margins and interiors (Schlager,
1993; Purdy & Winterer, 2001; Gischler, 2015). The degree of karstification is dictated
by subsidence rates and precipitation, which differs among platforms (Isaack & Gischler,
2015). The distinctive antecedent topography of each platform creates a unique landscape
for sediment accumulation. Though there is no data to reveal the structure and subsequent
influence of antecedent topography on the three study sites, inferences can be made from
observations of lagoonal depth and the topographic highs and lows on the platform tops
of Raivavae, Tubuai, and Bora Bora (Isaack & Gischler, 2015).
The combined effects of wind and wave energy, platform morphology and
orientation, and antecedent topography influence platform currents. Platform currents that
are strengthened by wind/wave activity, tides, and/or connectivity to stronger, open ocean
currents can be strong enough to transport sediments in the platform margin and interior
(Kench, 1998a; Kench, 1998b; Gischler, 2006; Rankey & Reeder, 2010; Harris et al.,
2014a). No data for the platform currents of the study sites was used, but inferences can
be made from spatial patterns of sediment grain size and from close inspection of satellite
imagery of each site. Gischler (2011) reports a high abundance (mean = 70%) of fines (s
< 0.125 mm) in the interior of Bora Bora interpreted to be derived from the breakdown of
skeletal fragments. Conversely, mud was ubiquitously low in the interior of Raivavae and
Tubuai. The lack of mud on Raivavae and Tubuai indicates a lack of production of fine
grains and/or a high-energy environmental setting. Nothing can be said regarding the
production of mud on Raivavae and Tubuai; however, there are hints of a high-energy
lagoonal environment on both platforms, which could account for the similarities in grain
texture and faunal grain type between the interior and margin environments of these two
sites.
Raivavae and Tubuai both have a prominent pass in the northern reef rim and
smaller passes in the southern reef rim that allow for improved hydrodynamic
connectivity between the open-ocean and lagoon. Close inspection of satellite imagery
for Raivavae and Tubuai reveals markings in the lagoon floor indicative of current flow
around the central island of each platform (Figs. 16 and 17). It is highly likely that the
currents in the lagoon of Raivavae and Tubuai are strong enough to winnow mud from
the lagoon. Gischler (2011) states that lagoonal circulation on Bora Bora is characterized
by water entering the lagoon via reef spill over and exiting through the Ava Nui channel
in the west. The lagoon of Bora Bora is substantially deeper than that of Raivavae and
Tubuai (depth: Bora Bora ~ 40 m and Raivavae and Tubuai ~ 20 m), and is broken up
into six basins. The reduced circulation caused by a combination of minimal lagoonal
input of open ocean water and a deep, disjointed lagoon creates a tranquil lagoonal
environment virtually devoid of strong currents and thus an ideal setting for the
accumulation of fine grains.
Figure 16: WorldView-2 satellite imagery of Raivavae. Subscene (A) provides a close up
highlighting sedimentary bedforms on the lagoon floor that may be indicative of current
flow around the central island of Raivavae. The lines with double arrows illustrate the
possible current directions around the central island as interpreted from the satellite
imagery.
Figure 17: WorldView-2 satellite imagery of Tubuai. Subscenes (A, B, and C) provide a
close ups showing sedimentary bedforms on the lagoon floor that may be indicative of
current flow around the central island of Tubuai. The lines with double arrows illustrate
the possible current directions around the central island as interpreted from the satellite
imagery.
Another major effect that contributes to the complexity of sediment production,
transportation and re-distribution is the assemblages of grain producers. In addition to
being produced on the platform margin, carbonate sediment can be produced in situ
within the back reef sand apron and lagoon. Major contributors to carbonate production
in these environments include: coral patch reefs, bivalve and gastropod fragments,
Halimeda plates and other calcareous green algae, and certain varieties of forams (Adjas
et al., 1990; Yamano et al., 2002; Gischler, 2011). In situ production from these
contributors directly impacts the grain texture and type by adding non-reef derived
sediment that is not in equilibrium with the surrounding ambient hydrodynamics
(Wasserman & Rankey, 2014). Close observations of satellite imagery available for the
three study sites shows that the lagoons of Raivavae and Tubuai have a substantial
number of patch reefs within the back reef apron and lagoon; however, Gischler (2011)
reports a scarcity of patch reefs in the lagoon of Bora Bora. It is highly plausible that in
situ production of sediment from lagoonal patch reefs and other sediment producers
coupled with differential patterns of hydrodynamics, influenced by local wave patterns
and platform morphology and orientation, are the main forces behind the unique patterns
of sediment production, redistribution and accumulation apparent on each study site. The
unique patterns, inherent to each platform, help explain the reduced level of accuracy and
the among and within platform differences observed in the water depth, RDRR, and LDA
models
7.2 Abundance of Coral Fragments: Indicator of Distance from Reef Rim with
Applications to the Rock Record
Results from Raivavae and Bora Bora show that the abundance of coral holds
potential to be an indicator of distance from the reef rim. In addition to being a variable in
the top selected water depth and RDRR models for Raivavae, coral was a reoccurring
variable in nine of the top 15 (five for each subgroup) water depth models and 10 of the
top 15 (five for each subgroup) RDRR models for Raivavae. On these study sites water
depth generally increases with increasing distance from the reef rim. Consequently, the
relationship between coral and water depth as seen in the models also has impactions to
distance from the reef rim. The recurrence of coral in the top Raivavae models and the
accuracy of the top selected Raivavae and Bora Bora water depth and RDRR models
suggest that coral is a primary feature linked to RDRR.
The zonation of coral reefs and their associated detritus make the abundance of
coral a suitable proxy to estimate the distance from the reef rim (Rankey et al., 2011;
Wasserman & Rankey, 2014). Coral grows in highest abundance on the platform margin.
Here, wave energy stirs up nutrients and flushes out waste and suspended sediment,
providing the ideal environment for corals (Scholle & Ulmer-Scholle, 2003; Schlager &
Purkis, 2013). Carbonate grains are generally deposited in close proximity to their
environment of origin, consequently a high abundance of coral fragments can be found
on the platform margin and back reef sand apron (Masse et al., 1989; Gischler &
Lomando, 1999; Gischler, 2006, 2011). Further lagoonward, wave energy dissipates and
depth increases; causing water quality and clarity to decrease and become less favorable
for coral. In result, the abundance of coral declines and is reflected in the surrounding
sediment (Ginsburg, 1956; Purdy, 1963; Milliman, 1967; Gischler & Lomando, 1999).
Lagoonal patch reefs are an exception. They contribute coral fragments to lagoonal
sediments, but the abundance of coral fragments dissipates with distance from patch reefs
(Tudhope et al., 1985). The presence of patch reefs provides variability in the trend
between distance from the reef rim and abundance of coral fragments; however, coral
rich sediments can still be considered generally reliable indicators of proximity to the reef
rim.
The relationship between distance from the reef rim and abundance of coral also
has implications to the rock record. Modern scleractinian corals have been major
contributors to reef framework, and thus carbonate platforms, since the Miocene (~ 23
MA) (Scholle & Ulmer-Scholle, 2003; Flügel, 2004). Ancient carbonate platforms are of
particular interest given their propensity to form excellent water aquifers and
hydrocarbon reservoirs. Sampling ancient platforms to describe their dimensional
anatomy and locate reservoirs is a difficult task that often involves interpreting lateral
facies heterogeneity from drilled rock cores collected from sparsely spaced wells, or from
analyzing vertical and horizontal facies changes within an outcrop. The relationship
between distance from the reef rim and coral abundance, as seen from the modern
examples presented in this study, may provide a tool to better interpret lateral facies
heterogeneity and predict locations of potential Miocene-age reservoirs based on core
samples.
8. Conclusion
In summary, this study revealed that on Raivavae, Tubuai, and Bora Bora,
sediment texture and type can be used to predict water depth and distance from the reef
rim and differentiate between marginal and interior environments with moderate accuracy
(≥ 65%). Among and within platform differences observed in water depth and RDRR
models prevented the application of a single model to more than one platform. The
exception being the Raivavae water depth model that predicted water depth on a
platform-wide scale based on the abundance of mud and coral. This model had an
accuracy of 81% and 78% when applied to Raivavae and Tubuai, respectively. The
overall reduced level of model accuracy and the among and within platform differences
of the models were likely due to several local effects that influence sediment production,
redistribution, and accumulation that produced unique sedimentary patterns among and
within platforms. Overall, the results of this study suggest that the abundance of coral
holds potential to be utilized as a proxy for distance from the reef rim on modern and
ancient isolated carbonate platforms.
To expand the applicability of the modeling performed in this study, future
studies should incorporate data from a larger swath of isolated platforms. Ideally, a more
thorough sampling scheme should be implemented as well. Sampling transects should
have multiple sediment sample locations and span from the reef rim to central lagoon.
The entirety of the platform should be sampled if possible. Investigations into
windward/swell-ward and leeward differences and analyzing models that capture these
differences could be performed with a more thoroughly sampled dataset.
9. References
Adjas, A., Masse, J.-P., & Montaggioni, L. F. (1990). Fine-grained carbonates in nearly
closed reef environments: Mataiva and Takapoto atolls, Central Pacific Ocean.
Sedimentary Geology, 67(1), 115-132.
Babyak, M. A. (2004). What you see may not be what you get: a brief, nontechnical
introduction to overfitting in regression-type models. Psychosomatic medicine,
66(3), 411-421.
Betzler, C., Lindhorst, S., Eberli, G. P., Lüdmann, T., Möbius, J., Ludwig, J., . . .
Hübscher, C. (2014). Periplatform drift: The combined result of contour current
and off-bank transport along carbonate platforms. Geology, 42(10), 871-874.
Blott, S. J., & Pye, K. (2001). GRADISTAT: a grain size distribution and statistics
package for the analysis of unconsolidated sediments. Earth surface processes
and Landforms, 26(11), 1237-1248.
Bosence, D. (2008). Randomness or order in the occurrence and preservation of shallow-
marine carbonate facies? Holocene, South Florida. Palaeogeography,
Palaeoclimatology, Palaeoecology, 270(3), 339-348.
Braga, J. C., Martín, J. M., & Riding, R. (1996). Internal structure of segment reefs:
Halimeda algal mounds in the Mediterranean Miocene. Geology, 24(1), 35-38.
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multimodel inference: a
practical information-theoretic approach: Springer.
Chevillon, C. (1996). Skeletal composition of modern lagoon sediments in New
Caledonia: coral, a minor constituent. Coral Reefs, 15(3), 199-207.
Eberli, G. P. (1989). Cenozoic progradation of northwestern Great Bahama Bank, a
record of lateral platform growth and sea-level fluctuations.
Fisher, R. A. (1936). The use of multiple measurements in taxonomic problems. Annals
of eugenics, 7(2), 179-188.
Flügel, E. (2004). Microfacies of carbonate rocks: analysis, interpretation and
application: Springer.
Folk, R. L., & Ward, W. C. (1957). Brazos River bar: a study in the significance of grain
size parameters. Journal of Sedimentary Research, 27(1).
Ginsburg, R. N. (1956). Environmental relationships of grain size and constituent
particles in some south Florida carbonate sediments. AAPG bulletin, 40(10),
2384-2427.
Gischler, E. (2006). Sedimentation on Rasdhoo and Ari Atolls, Maldives, Indian Ocean.
Facies, 52(3), 341-360.
Gischler, E. (2011). Sedimentary facies of Bora Bora, Darwin's type barrier reef (Society
Islands, south Pacific): the unexpected occurrence of non-skeletal grains. Journal
of Sedimentary Research, 81(1), 1-17.
Gischler, E. (2015). Quaternary reef response to sea‐level and environmental change in
the western Atlantic. Sedimentology, 62(2), 429-465.
Gischler, E., & Lomando, A. J. (1999). Recent sedimentary facies of isolated carbonate
platforms, Belize-Yucatan system, Central America. Journal of Sedimentary
Research, 69(3).
Gischler, E., & Zingeler, D. (2002). The origin of carbonate mud in isolated carbonate
platforms of Belize, Central America. International Journal of Earth Sciences,
91(6), 1054-1070.
Harris, D. L., Vila-Concejo, A., & Webster, J. M. (2014a). Geomorphology and sediment
transport on a submerged back-reef sand apron: One Tree Reef, Great Barrier
Reef. Geomorphology, 222, 132-142.
Harris, P., Purkis, S., Ellis, J., Swart, P., & Reijmer, J. (2014b). Mapping water-depth and
depositional facies on Great Bahama Bank. Sedimentology.
Harris, P., & Vlaswinkel, B. (2008). Modern isolated carbonate platforms: Templates for
quantifying facies attributes of hydrocarbon reservoirs. Controls on carbonate
platform and reef development: SEPM Special Publication, 89, 323-341.
Harris, P. M. (1979). Facies anatomy and diagenesis of a Bahamian ooid shoal (Vol. 7):
Comparative Sedimentology Laboratory, Division of Marine Geology and
Geophysics, University of Miami, Rosenstiel School of Marine & Atmospheric
Science.
Harris, P. M. (1983). The Joulters ooid shoal, Great Bahama Bank Coated Grains (pp.
132-141): Springer.
Harris, P. M. M., Purkis, S. J., & Ellis, J. (2011). Analyzing spatial patterns in modern
carbonate sand bodies from Great Bahama Bank. Journal of Sedimentary
Research, 81(3), 185-206.
Hillis-Colinvaux, L. (1980). Ecology and taxonomy of Halimeda: primary producer of
coral reefs. Advances in marine biology, 17, 1-327.
Hine, A. C., Wilber, R. J., & Neumann, A. C. (1981). Carbonate sand bodies along
contrasting shallow bank margins facing open seaways in northern Bahamas.
AAPG bulletin, 65(2), 261-290.
Illing, L. V. (1954). Bahaman calcareous sands. AAPG bulletin, 38(1), 1-95.
Imbrie, J., & Purdy, E. G. (1962). Classification of modern Bahamian carbonate
sediments.
Isaack, A., & Gischler, E. (2015). The significance of sand aprons in Holocene atolls and
carbonate platforms. Carbonates and Evaporites, 1-13.
Kaczmarek, S. E., Hicks, M. K., Fullmer, S. M., Steffen, K. L., & Bachtel, S. L. (2010).
Mapping facies distributions on modern carbonate platforms through integration
of multispectral Landsat data, statistics-based unsupervised classifications, and
surface sediment data. AAPG bulletin, 94(10), 1581-1606.
Kench, P. (1998a). A currents of removal approach for interpreting carbonate
sedimentary processes. Marine Geology, 145(3), 197-223.
Kench, P. S. (1998b). Physical controls on development of lagoon sand deposits and
lagoon infilling in an Indian Ocean atoll. Journal of Coastal Research, 1014-
1024.
Kohler, K. E., & Gill, S. M. (2006). Coral Point Count with Excel extensions (CPCe): A
Visual Basic program for the determination of coral and substrate coverage using
random point count methodology. Computers & Geosciences, 32(9), 1259-1269.
Lachenbruch, P. A., & Mickey, M. R. (1968). Estimation of error rates in discriminant
analysis. Technometrics, 10(1), 1-11.
Madden, R. H., Wilson, M. E., & O'Shea, M. (2013). Modern fringing reef carbonates
from equatorial SE Asia: An integrated environmental, sediment and satellite
characterisation study. Marine Geology, 344, 163-185.
Masse, J., Thomassin, B., & Acquaviva, M. (1989). Bioclastic sedimentary environments
of coral reefs and lagoon around Mayotte island (Comoro archipelago,
Mozambique channel, SW Indian Ocean). Journal of Coastal Research, 419-432.
McLachlan, G. (2004). Discriminant analysis and statistical pattern recognition (Vol.
544): John Wiley & Sons.
Milliman, J. D. (1967). Carbonate sedimentation on Hogsty Reef, a Bahamian atoll.
Journal of Sedimentary Research, 37(2).
Montaggioni, L. F. (2005). History of Indo-Pacific coral reef systems since the last
glaciation: development patterns and controlling factors. Earth-Science Reviews,
71(1), 1-75.
Purdy, E. G. (1963). Recent calcium carbonate facies of the Great Bahama Bank. 2.
Sedimentary facies. The Journal of Geology, 472-497.
Purdy, E. G., & Winterer, E. L. (2001). Origin of atoll lagoons. Geological Society of
America Bulletin, 113(7), 837-854.
Purkis, S., Casini, G., Hunt, D., & Colpaert, A. (2015a). Morphometric patterns in
Modern carbonate platforms can be applied to the ancient rock record:
Similarities between Modern Alacranes Reef and Upper Palaeozoic platforms of
the Barents Sea. Sedimentary Geology, 321, 49-69.
Purkis, S., Kerr, J., Dempsey, A., Calhoun, A., Metsamaa, L., Riegl, B., . . . Renaud, P.
(2014). Large-scale carbonate platform development of Cay Sal Bank, Bahamas,
and implications for associated reef geomorphology. Geomorphology.
Purkis, S., Vlaswinkel, B., & Gracias, N. (2012a). Vertical-To-Lateral Transitions
Among Cretaceous Carbonate Facies—A Means To 3-D Framework Construction
Via Markov Analysis. Journal of Sedimentary Research, 82(4), 232-243.
Purkis, S. J., & Harris, P. M. M. (2016). The Extent and Patterns of Sediment Filling of
Accommodation Space On Great Bahama Bank. Journal of Sedimentary
Research, 86(4), 294-310.
Purkis, S. J., Harris, P. M. M., & Ellis, J. (2012b). Patterns of sedimentation in the
contemporary Red Sea as an analog for ancient carbonates in rift settings. Journal
of Sedimentary Research, 82(11), 859-870.
Purkis, S. J., Kohler, K. E., Riegl, B. M., & Rohmann, S. O. (2007). The statistics of
natural shapes in modern coral reef landscapes. The Journal of Geology, 115(5),
493-508.
Purkis, S. J., Riegl, B. M., & Andréfouët, S. (2005). Remote sensing of geomorphology
and facies patterns on a modern carbonate ramp (Arabian Gulf, Dubai, UAE).
Journal of Sedimentary Research, 75(5), 861-876.
Purkis, S. J., Rowlands, G. P., & Kerr, J. M. (2015b). Unravelling the influence of water
depth and wave energy on the facies diversity of shelf carbonates. Sedimentology.
Purkis, S. J., & Vlaswinkel, B. (2012). Visualizing lateral anisotropy in modern
carbonates. AAPG bulletin, 96(9), 1665-1685.
Rankey, E. C. (2004). On the interpretation of shallow shelf carbonate facies and
habitats: how much does water depth matter? Journal of Sedimentary Research,
74(1), 2-6.
Rankey, E. C., & Garza-Pérez, J. R. (2012). Seascape metrics of shelf-margin reefs and
reef sand aprons of Holocene carbonate platforms. Journal of Sedimentary
Research, 82(1), 57-75.
Rankey, E. C., Guidry, S. A., Reeder, S. L., & Guarin, H. (2009). Geomorphic and
sedimentologic heterogeneity along a Holocene shelf margin: Caicos Platform.
Journal of Sedimentary Research, 79(6), 440-456.
Rankey, E. C., & Reeder, S. L. (2010). Controls on platform‐scale patterns of surface
sediments, shallow Holocene platforms, Bahamas. Sedimentology, 57(6), 1545-
1565.
Rankey, E. C., & Reeder, S. L. (2011). Holocene oolitic marine sand complexes of the
Bahamas. Journal of Sedimentary Research, 81(2), 97-117.
Rankey, E. C., Reeder, S. L., & Garza-Pérez, J. R. (2011). Controls On Links Between
Geomorphical and Surface Sedimentological Variability: Aitutaki and Maupiti
Atolls, South Pacific Ocean. Journal of Sedimentary Research, 81(12), 885-900.
Rees, S. A., Opdyke, B. N., Wilson, P. A., & Fifield, L. K. (2005). Coral reef
sedimentation on Rodrigues and the Western Indian Ocean and its impact on the
carbon cycle. Philosophical Transactions of the Royal Society A: Mathematical,
Physical and Engineering Sciences, 363(1826), 101-120.
Reijmer, J. J., Swart, P. K., Bauch, T., Otto, R., Reuning, L., Roth, S., & Zechel, S.
(2012). A Re‐Evaluation of Facies on Great Bahama Bank I: New Facies Maps
of Western Great Bahama Bank. Perspectives in Carbonate Geology: A Tribute to
the Career of Robert Nathan Ginsburg (Special Publication 41 of the IAS), 98,
29-46.
Riegl, B., Halfar, J., Purkis, S., & Godinez-Orta, L. (2007). Sedimentary facies of the
eastern Pacific's northernmost reef-like setting (Cabo Pulmo, Mexico). Marine
Geology, 236(1), 61-77.
Schlager, W. (1993). Accommodation and supply—a dual control on stratigraphic
sequences. Sedimentary Geology, 86(1), 111-136.
Schlager, W. (2007). Carbonate Sedimentology and Sequence Stratigraphy (Vol. 8):
SEPM Soc for Sed Geology.
Schlager, W., & Purkis, S. J. (2013). Bucket structure in carbonate accumulations of the
Maldive, Chagos and Laccadive archipelagos. International Journal of Earth
Sciences, 102(8), 2225-2238.
Scholle, P. A., & Ulmer-Scholle, D. S. (2003). A Color Guide to the Petrography of
Carbonate Rocks: Grains, Textures, Porosity, Diagenesis, AAPG Memoir 77
(Vol. 77): AAPG.
Team, R. C. (2014). R: A Language and Environment for Statistical Computing.
Triffleman, N. J., Hallock, P., & Hine, A. C. (1992). Morphology, sediments, and
depositional environments of a small carbonate platform; Serranilla Bank,
Nicaraguan Rise, Southwest Caribbean Sea. Journal of Sedimentary Research,
62(4), 591-606.
Tucker, M. E., & Wright, V. P. (2009). Carbonate sedimentology: John Wiley & Sons.
Tudhope, A., Scoffin, T., Stoddart, D., & Woodroffe, C. (1985). Sediments of Suwarrow
atoll. Paper presented at the Proc 5th Int Coral Reef Symp.
Wasserman, H. N., & Rankey, E. C. (2014). Physical Oceanographic Influences On
Sedimentology of Reef Sand Aprons: Holocene of Aranuka Atoll (Kiribati),
Equatorial Pacific. Journal of Sedimentary Research, 84(7), 586-604.
Weber, J. N., & Woodhead, P. M. (1972). Carbonate lagoon and beach sediments of
Tarawa Atoll, Gilbert Islands: Smithsonian Institution.
Wetcher-Hendricks, D. (2011). Analyzing quantitative data: An introduction for social
researchers: John Wiley & Sons.
Wilber, R. J., Milliman, J. D., & Halley, R. B. (1990). Accumulation of bank-top
sediment on the western slope of Great Bahama Bank: rapid progradation of a
carbonate megabank. Geology, 18(10), 970-974.
Wisuki. (2012a). Bora-Bora Statistics. Retrieved 2013, 1 Jun, from
http://wisuki.com/statistics/3883/bora-bora
Wisuki. (2012b). Tubuai Statistics. Retrieved 1 Jun, 2013, from
http://wisuki.com/statistics/3900/tubuai.
Wright, V. P., & Burgess, P. M. (2005). The carbonate factory continuum, facies mosaics
and microfacies: an appraisal of some of the key concepts underpinning carbonate
sedimentology. Facies, 51(1-4), 17-23.
Yamano, H., Kayanne, H., Matsuda, F., & Tsuji, Y. (2002). Lagoonal facies, ages, and
sedimentation in three atolls in the Pacific. Marine Geology, 185(3), 233-247.
Appendix A: Photolog of Faunal Grain Types
Bryozoan (thin section): colonial with
many small, boxy pores, or zooecial
apertures, where bryozoan polyps once
lived. The main difference between
bryozoans and coral are smaller colony
size and individual living chambers
(Scholle and Ulmer-Scholle 2003).
Bryozoan (loose grain): bryozoans have a
similar porous external structure to corals,
but at a smaller scale. Compare this
picture with the picture of loose coral
grains below.
Coral (thin section): colonial with many
pores, corallites, where coral polyps once
lived. The yellow white is the coral
skeleton while the brown is fine grained
mud, or matrix, that filled the interstitial
space between the skeleton.
Coral (loose grains): coral fragments may
be in many forms: from branching (top
left) to more mound like (other grains)
each with numerous corallites. Horizontal
tabulae and vertical corallite walls create
the cell-like internal structure of the coral
skeleton.
Coralline Red Algae (thin section): brown
grain with fine scale reticulate latticework
internal structure that reflects the
filamentous morphology of this algae
(Scholle and Ulmer-Scholle 2003).
Coralline Red Algae (loose grain): no
example
Halimeda (thin section): brown grain
porous internal structure. The internal
structure is a result of complexly
intertwined filaments (Scholle and Ulmer-
Scholle 2003).
Halimeda (thin section): larger club
shaped grain. The outside layer of the
grain may be smooth or covered in tiny
pores, or utricles. The internal structure is
heavily reticulated.
Mollusk (thin section): gastropods (above)
are noted by their strongly curved and
smooth shape. Larger, whole specimens
have a spiral shape. Bivalves (not
pictured) typically appear as thin slightly
curved fragments.
Mollusk (loose grains): gastropods (top
left and bottom) and mollusks (top right)
are easily identifiable. Gastropods have a
spiral structure while bivalves appear as
single slightly curved shell fragments.
Foram (thin section): most foram tests,
outer shell, (boxed in red) are
multichambered and can resemble a
mollusk shell. However, foram tests are
usually much smaller than mollusk shells.
Compare scale of above photo.
Foram (loose grains): small tests, usually
with a shiny appearance. There are a
variety of test morphologies. Above are
the most common types found from
Raivavae and Tubuai. Tests of these
forams are small and flat with a circular
shape. Most appear as full tests.
Echinoderm (thin section): echinoderm
spines (above) and plates (not pictured)
are perforated with a meshwork of
honeycomb-like pores. Transversely cut
spines are circular with radial symmetry,
while a longitudinal sections show long,
slender spines with a striped pattern of
pore space. The most distinguishing
characteristic of echinoderms is their
visible extinction of internal structures
when viewed with cross polarized light
(Scholle and Ulmer-Scholle 2003).
Echinoderm (loose grains): appear as
spines (top and bottom) or plates (center)
Spines are long and slender and often
have a striped pattern. The base of a spine
has a collar followed by a ball-like
structure that attaches the spine to the
plate. The plate (center) shows the spine
attachment point as a prominent ball-like
structure. Echinoderm plates have a very
fine porous structure.
Spicule (thin section): no example
Spicule (loose grains): calcareous
structures found in the tissue of
octocorals. Spicules are small spindle-
shaped rods with pointed ends and small
protrusions or spines (Scholle and Ulmer-
Scholle 2003).
Serpulid (thin section): no example
Serpulid (loose grain): small slender tube
typically encrusting upon other grains, but
can be found as a solitary structure as well
(above).
Crustacean (thin section): no example
Crustacean (loose grains): fragments of
the crustacean exoskeleton. Above are the
appendages from a crab exoskeleton.
Notable by jointed appendages.
Appendix B: Referenced Tables
Table 1: Sample ID, location, in terms of (margin or interior), RDRR, water depth,
percent gravel, percent sand, percent mud, and sorting values of each of the sediment
samples collected from the platform top of Raivavae.
Raivavae
Sample Location RDRR Depth (m) Gravel Sand Mud Sorting
FPA-8 Interior 0.27 4.21 14.00% 79.00% 7.00% 3.39
FPA-17 Interior 0.33 3.52 6.00% 90.00% 4.00% 2.46
FPA-16 Interior 0.35 4.41 15.00% 83.00% 2.00% 2.79
FPA-11 Interior 0.4 6.48 4.00% 92.00% 5.00% 1.98
FPA-20 Interior 0.49 4.86 8.00% 83.00% 9.00% 2.65
FPA-26 Interior 0.49 7.76 27.00% 59.00% 15.00% 3.08
FPA-22 Interior 0.64 8.21 20.00% 74.00% 7.00% 3.5
FPA-19 Interior 0.67 7.06 5.00% 52.00% 43.00% 3.18
FPA-6 Interior 0.67 4.76 16.00% 83.00% 1.00% 2.93
FPA-28 Interior 0.71 8.28 11.00% 88.00% 1.00% 3.13
FPA-13 Interior 0.76 5.18 13.00% 87.00% 0.00% 2.21
FPA-29 Interior 0.76 10.79 13.00% 82.00% 4.00% 3.17
FPA-2 Interior 0.82 14.88 6.00% 90.00% 3.00% 2.46
FPA-10 Interior 0.89 6.8 20.00% 76.00% 3.00% 2.94
FPA-9 Margin 0.17 2.56 30.00% 70.00% 0.00% 2.01
FPA-24 Margin 0.2 1.23 23.00% 77.00% 0.00% 2.99
FPA-25 Margin 0.21 0.72 16.00% 84.00% 0.00% 3.04
FPA-1 Margin 0.29 4.37 3.00% 97.00% 0.00% 1.59
FPA-18 Margin 0.34 1.48 33.00% 67.00% 0.00% 3.03
FPA-21 Margin 0.36 1.42 13.00% 87.00% 0.00% 2.19
FPA-12 Margin 0.39 1.36 17.00% 82.00% 0.00% 2.6
FPA-30 Margin 0.43 1.98 5.00% 92.00% 2.00% 3.21
FPA-15 Margin 0.44 1.22 12.00% 88.00% 0.00% 2
FPA-14 Margin 0.56 1.55 14.00% 85.00% 0.00% 2.26
FPA-23 Margin 0.59 1.97 10.00% 90.00% 0.00% 2.95
FPA-3 Margin 0.72 1.6 6.00% 94.00% 0.00% 2.21
FPA-4 Margin 0.72 1.6 13.00% 87.00% 0.00% 1.56
FPA-5 Margin 0.72 1.6 3.00% 97.00% 0.00% 1.57
Table 2: Sample ID, abundance of each faunal grain type observed in Raivavae sediment
samples. BR = bryozoan, CR = coral, CA = coralline algae, HA = Halimeda, MO =
mollusk, EC = echinoderm, FR = foram, SP = spicule, SR = serpulid, CR = crustacean, &
UN = unknown.
Raivavae
Sample BR CR CA CA HA MO EC FR SP SR CR UN
FPA-8 0.00% 40.00% 17.00% 57.00% 18.00% 14.00% 0.00% 0.00% 0.00% 0.00% 0.00% 12.00%
FPA-17 0.00% 23.00% 22.00% 45.00% 28.00% 12.00% 1.00% 5.00% 0.00% 0.00% 0.00% 9.00%
FPA-16 0.00% 23.00% 19.00% 42.00% 38.00% 15.00% 0.00% 0.00% 0.00% 0.00% 0.00% 5.00%
FPA-11 1.00% 25.00% 21.00% 46.00% 28.00% 5.00% 0.00% 10.00% 0.00% 1.00% 0.00% 10.00%
FPA-20 0.00% 27.00% 16.00% 43.00% 19.00% 23.00% 0.00% 2.00% 0.00% 0.00% 0.00% 13.00%
FPA-26 0.00% 12.00% 0.00% 12.00% 63.00% 21.00% 0.00% 0.00% 0.00% 0.00% 0.00% 4.00%
FPA-22 1.00% 19.00% 11.00% 30.00% 42.00% 19.00% 0.00% 1.00% 0.00% 0.00% 0.00% 7.00%
FPA-19 0.00% 24.00% 1.00% 25.00% 56.00% 14.00% 0.00% 0.00% 0.00% 0.00% 0.00% 5.00%
FPA-6 0.00% 20.00% 8.00% 28.00% 69.00% 3.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00%
FPA-28 0.00% 34.00% 6.00% 40.00% 16.00% 35.00% 0.00% 2.00% 1.00% 0.00% 0.00% 6.00%
FPA-13 0.00% 0.00% 0.00% 0.00% 71.00% 22.00% 0.00% 1.00% 0.00% 0.00% 0.00% 6.00%
FPA-29 2.00% 13.00% 3.00% 16.00% 51.00% 12.00% 0.00% 10.00% 0.00% 0.00% 0.00% 9.00%
FPA-2 0.00% 8.00% 0.00% 8.00% 58.00% 22.00% 1.00% 2.00% 0.00% 0.00% 0.00% 9.00%
FPA-10 1.00% 10.00% 0.00% 10.00% 56.00% 25.00% 0.00% 0.00% 0.00% 3.00% 0.00% 5.00%
FPA-9 0.00% 44.00% 4.00% 48.00% 19.00% 27.00% 2.00% 0.00% 1.00% 0.00% 0.00% 3.00%
FPA-24 0.00% 30.00% 14.00% 44.00% 33.00% 11.00% 6.00% 0.00% 1.00% 0.00% 0.00% 5.00%
FPA-25 0.00% 32.00% 9.00% 41.00% 31.00% 17.00% 1.00% 0.00% 2.00% 1.00% 0.00% 7.00%
FPA-1 1.00% 46.00% 27.00% 73.00% 15.00% 3.00% 2.00% 0.00% 0.00% 0.00% 0.00% 6.00%
FPA-18 0.00% 32.00% 10.00% 42.00% 26.00% 30.00% 1.00% 0.00% 1.00% 1.00% 0.00% 0.00%
FPA-21 0.00% 38.00% 4.00% 42.00% 22.00% 29.00% 0.00% 0.00% 0.00% 2.00% 0.00% 5.00%
FPA-12 0.00% 25.00% 15.00% 40.00% 26.00% 20.00% 5.00% 0.00% 2.00% 0.00% 0.00% 7.00%
FPA-30 0.00% 33.00% 19.00% 52.00% 12.00% 26.00% 0.00% 2.00% 0.00% 0.00% 0.00% 8.00%
FPA-15 0.00% 39.00% 6.00% 45.00% 22.00% 22.00% 2.00% 0.00% 6.00% 0.00% 0.00% 3.00%
FPA-14 0.00% 24.00% 11.00% 35.00% 45.00% 11.00% 4.00% 0.00% 0.00% 1.00% 0.00% 4.00%
FPA-23 0.00% 42.00% 8.00% 50.00% 16.00% 23.00% 0.00% 1.00% 0.00% 1.00% 0.00% 10.00%
FPA-3 0.00% 17.00% 11.00% 28.00% 27.00% 30.00% 0.00% 0.00% 1.00% 0.00% 0.00% 14.00%
FPA-4 0.00% 43.00% 3.00% 46.00% 18.00% 32.00% 0.00% 0.00% 0.00% 0.00% 0.00% 4.00%
FPA-5 0.00% 39.00% 14.00% 53.00% 16.00% 21.00% 3.00% 0.00% 1.00% 0.00% 0.00% 6.00%
Table 3: Sample ID, location, in terms of (Margin or Interior), RDRR, water depth,
percent gravel, percent sand, percent mud, and sorting values of each of the sediment
samples collected from the platform top of Tubuai.
Tubuai
Sample Location RDRR Depth (m) Gravel Sand Mud Sorting
FPA-70 Interior 0.45 21.76 5.00% 60.00% 35.00% 1.4
FPA-45 Interior 0.57 5.75 18.00% 82.00% 0.00% 1.32
FPA-57 Interior 0.59 15.94 5.00% 77.00% 18.00% 1.23
FPA-58 Interior 0.65 4.28 4.00% 96.00% 0.00% 0.85
FPA-55 Interior 0.77 1.64 28.00% 70.00% 2.00% 1.48
FPA-56 Interior 0.78 7.28 12.00% 85.00% 3.00% 1.59
FPA-60 Margin 0.09 5.31 17.00% 83.00% 0.00% 0.86
FPA-61 Margin 0.09 5.31 6.00% 94.00% 0.00% 0.87
FPA-59 Margin 0.12 12.11 10.00% 89.00% 0.00% 1.23
FPA-71 Margin 0.16 1.62 23.00% 75.00% 1.00% 1.38
FPA-69 Margin 0.2 0.74 63.00% 38.00% 0.00% 1.15
FPA-68 Margin 0.22 1.38 36.00% 64.00% 0.00% 1.2
FPA-54 Margin 0.47 1.8 36.00% 64.00% 0.00% 1.51
FPA-49 Margin 0.49 1.35 27.00% 72.00% 0.00% 0.91
FPA-53 Margin 0.6 1.37 18.00% 82.00% 0.00% 1.19
FPA-46 Margin 0.64 5.58 18.00% 82.00% 0.00% 1.05
FPA-48 Margin 0.64 1.89 18.00% 81.00% 0.00% 1.4
FPA-47 Margin 0.78 5.01 20.00% 79.00% 0.00% 1.52
FPA-50 Margin 0.81 3.29 23.00% 77.00% 0.00% 1.16
FPA-52 Margin 0.81 1.42 8.00% 92.00% 0.00% 1.05
FPA-51 Margin 0.86 2.69 30.00% 69.00% 0.00% 1.09
Table 4: Sample ID, abundance of each faunal grain type observed in Tubuai sediment
samples. BR = bryozoan, CR = coral, CA = coralline algae, HA = Halimeda, MO =
mollusk, EC = echinoderm, FR = foram, SP = spicule, SR = serpulid, CR = crustacean, &
UN = unknown.
Tubuai
Sample BR CR CA CA HA MO EC FR SP SR CR UN
FPA-70 0.00% 21.00% 9.00% 30.00% 35.00% 27.00% 0.00% 0.00% 0.00% 0.00% 0.00% 8.00%
FPA-45 0.00% 36.00% 13.00% 49.00% 13.00% 26.00% 0.00% 0.00% 3.00% 0.00% 0.00% 9.00%
FPA-57 0.00% 20.00% 18.00% 38.00% 25.00% 25.00% 0.00% 0.00% 0.00% 0.00% 0.00% 12.00%
FPA-58 0.00% 20.00% 15.00% 35.00% 25.00% 26.00% 2.00% 1.00% 3.00% 1.00% 0.00% 7.00%
FPA-55 0.00% 8.00% 5.00% 13.00% 59.00% 20.00% 0.00% 4.00% 0.00% 0.00% 0.00% 4.00%
FPA-56 0.00% 20.00% 14.00% 34.00% 41.00% 16.00% 1.00% 3.00% 0.00% 0.00% 0.00% 5.00%
FPA-60 0.00% 23.00% 13.00% 36.00% 26.00% 24.00% 5.00% 2.00% 1.00% 0.00% 0.00% 6.00%
FPA-61 0.00% 30.00% 5.00% 35.00% 32.00% 22.00% 3.00% 2.00% 0.00% 0.00% 1.00% 4.00%
FPA-59 0.00% 23.00% 14.00% 37.00% 23.00% 20.00% 6.00% 1.00% 1.00% 0.00% 2.00% 10.00%
FPA-71 0.00% 20.00% 14.00% 34.00% 20.00% 32.00% 6.00% 0.00% 0.00% 0.00% 0.00% 8.00%
FPA-69 0.00% 14.00% 7.00% 21.00% 42.00% 24.00% 11.00% 0.00% 0.00% 0.00% 0.00% 2.00%
FPA-68 0.00% 12.00% 14.00% 26.00% 34.00% 22.00% 6.00% 0.00% 2.00% 0.00% 0.00% 10.00%
FPA-54 0.00% 25.00% 11.00% 36.00% 19.00% 35.00% 2.00% 0.00% 1.00% 0.00% 0.00% 7.00%
FPA-49 0.00% 35.00% 8.00% 43.00% 22.00% 25.00% 1.00% 0.00% 1.00% 0.00% 0.00% 8.00%
FPA-53 0.00% 17.00% 11.00% 28.00% 23.00% 40.00% 3.00% 0.00% 0.00% 0.00% 0.00% 6.00%
FPA-46 0.00% 37.00% 10.00% 47.00% 15.00% 29.00% 2.00% 0.00% 0.00% 0.00% 0.00% 7.00%
FPA-48 0.00% 35.00% 17.00% 52.00% 15.00% 15.00% 2.00% 0.00% 3.00% 1.00% 0.00% 12.00%
FPA-47 0.00% 25.00% 15.00% 39.00% 33.00% 18.00% 2.00% 1.00% 0.00% 2.00% 0.00% 6.00%
FPA-50 0.00% 27.00% 9.00% 36.00% 27.00% 27.00% 4.00% 0.00% 2.00% 1.00% 0.00% 3.00%
FPA-52 0.00% 21.00% 10.00% 31.00% 18.00% 37.00% 3.00% 0.00% 2.00% 0.00% 0.00% 9.00%
FPA-51 0.00% 25.00% 8.00% 33.00% 23.00% 35.00% 2.00% 0.00% 1.00% 0.00% 0.00% 6.00%
Table 5: Location, in terms of (margin or interior), RDRR, water depth, percent coral
(CR), and percent Halimeda (HA) of each of the sediment samples from the platform top
of Bora Bora. Sample ID, water depth, CR, and HA from (Gischler (2011)).
Bora Bora
Sample Location RDRR Depth (m) CR HA
BB-10 Interior 0.91 9 53.00% 4.60%
BB-14 Interior 0.54 24 25.50% 6.10%
BB-15 Interior 0.65 22 8.40% 10.70%
BB-16 Interior 0.82 24 1.40% 2.10%
BB-18 Interior 0.28 13 48.30% 7.90%
BB-19 Interior 0.35 22 18.70% 9.10%
BB-20 Interior 0.49 26 6.00% 1.50%
BB-21 Interior 0.49 28 0.10% 0.20%
BB-22 Interior 0.79 26 0.00% 0.00%
BB-23 Interior 0.83 22 0.00% 0.20%
BB-24 Interior 0.88 20 5.80% 4.20%
BB-25 Interior 0.96 16 10.70% 7.90%
BB-28 Interior 0.62 18 8.50% 9.60%
BB-30 Interior 0.74 30 0.10% 0.30%
BB-5 Interior 0.61 35 0.40% 1.70%
BB-6 Interior 0.65 37 0.60% 0.40%
BB-7 Interior 0.72 34 0.20% 0.50%
BB-8 Interior 0.78 30 6.70% 9.00%
BB-9 Interior 0.83 29 11.20% 31.90%
BB-1 Margin 0.22 0.5 2.00% 4.80%
BB-11 Margin 0.56 3 31.80% 23.30%
BB-12 Margin 0.61 2 28.50% 39.00%
BB-13 Margin 0.38 2 40.70% 24.20%
BB-17 Margin 0.23 3 32.30% 14.90%
BB-2 Margin 0.51 5 2.80% 12.00%
BB-26 Margin 0.45 0.5 28.20% 40.50%
BB-29 Margin 0.45 4 8.40% 6.30%
BB-3 Margin 0.73 11 3.70% 1.60%
BB-31 Margin 0.07 0 50.70% 10.20%
BB-4 Margin 0.07 25 28.80% 21.20%
Table 6: Results from all 22 Raivavae water depth models as applied to each of the three
subgroups of samples. RMSE is given in meters. NRMSE is the RMSE normalized to the
depth ranges of the respective subgroup. NRMSE gives the error (on a scale of 0 – 1) of
the model. Model structure represents the variables that were included in each model (1 =
randomness model, MU = mud, SA = sand, GR = gravel, SO = sorting, CR = coral, and
HA = Halimeda).
Water Depth (m)
Model #
Model
Structure
RMSE (Platform-
wide)
NRMSE (Platform-
wide)
RMSE
(Margin)
NRMSE
(Margin)
RMSE
(Interior)
NRMSE
(Interior)
0 1 3.37 0.2378 0.83 0.2277 2.93 0.2576
1 MU 3.34 0.2358 0.80 0.2197 2.92 0.2572
2 SA 3.31 0.2340 0.80 0.2194 2.93 0.2576
3 GR 3.16 0.2233 0.83 0.2271 2.93 0.2576
4 SO 3.21 0.2269 0.76 0.2075 2.92 0.2569
5 CR 2.74 0.1934 0.72 0.1975 2.69 0.2372
6 HA 2.83 0.2000 0.73 0.2013 2.81 0.2476
7 MU + SA 3.14 0.2220 0.80 0.2192 2.92 0.2571
8 MU + GR 3.15 0.2224 0.80 0.2197 2.92 0.2572
9 MU + SO 3.11 0.2195 0.75 0.2054 2.90 0.2552
10 MU + CR 2.69 0.1896 0.69 0.1895 2.65 0.2332
11 MU + HA 2.71 0.1914 0.73 0.1999 2.77 0.2436
12 SA + GR 3.15 0.2226 0.80 0.2193 2.93 0.2576
13 SA + SO 3.21 0.2268 0.75 0.2058 2.91 0.2565
14 SA + CR 2.73 0.1928 0.69 0.1900 2.69 0.2372
15 SA + HA 2.83 0.1996 0.73 0.1997 2.80 0.2462
16 GR + SO 3.09 0.2185 0.73 0.1987 2.92 0.2569
17 GR + CR 2.62 0.1850 0.72 0.1962 2.69 0.2365
18 GR + HA 2.77 0.1956 0.73 0.1998 2.81 0.2475
19 SO + CR 2.70 0.1907 0.69 0.1902 2.62 0.2303
20 SO + HA 2.78 0.1965 0.69 0.1883 2.80 0.2464
21 CR + HA 2.72 0.1921 0.70 0.1931 2.68 0.2362
Table 7: Results from all 22 Tubuai water depth models as applied to each of the three
subgroups of samples. NRMSE is the RMSE normalized to the depth ranges of the
respective subgroup. NRMSE gives the error (on a scale of 0 – 1) of the model. Model
structure represents the variables that were included in each model (1 = randomness
model, MU = mud, SA = sand, GR = gravel, SO = sorting, CR = coral, and HA =
Halimeda).
Water Depth (m)
Model #
Model
Structure
RMSE (Platform-
wide)
NRMSE (Platform-
wide)
RMSE
(Margin)
NRMSE
(Margin)
RMSE
(Interior)
NRMSE
(Interior)
0 1 5.25 0.2498 2.86 0.2513 7.07 0.3513
1 MU 4.36 0.2075 2.48 0.2178 5.41 0.2689
2 SA 5.25 0.2496 2.48 0.2178 5.67 0.2816
3 GR 2.68 0.1277 2.82 0.2478 1.98 0.0983
4 SO 5.19 0.2468 2.84 0.2502 7.03 0.3496
5 CR 5.25 0.2498 2.79 0.2452 7.00 0.3481
6 HA 5.24 0.2494 2.85 0.2510 6.90 0.3432
7 MU + SA 2.28 0.1085 2.48 0.2178 1.48 0.0734
8 MU + GR 2.27 0.1082 2.43 0.2135 1.48 0.0734
9 MU + SO 4.20 0.1996 2.48 0.2178 4.05 0.2013
10 MU + CR 4.31 0.2053 2.48 0.2178 5.41 0.2688
11 MU + HA 4.23 0.2010 2.38 0.2089 5.31 0.2638
12 SA + GR 2.28 0.1083 2.44 0.2143 1.48 0.0734
13 SA + SO 5.17 0.2458 2.48 0.2177 5.27 0.2620
14 SA + CR 5.25 0.2496 2.48 0.2178 5.26 0.2616
15 SA + HA 5.23 0.2489 2.38 0.2096 4.43 0.2200
16 GR + SO 2.67 0.1272 2.81 0.2475 1.95 0.0968
17 GR + CR 2.57 0.1220 2.76 0.2430 1.43 0.0711
18 GR + HA 2.63 0.1250 2.81 0.2470 1.26 0.0624
19 SO + CR 5.18 0.2465 2.78 0.2448 6.95 0.3454
20 SO + HA 5.19 0.2468 2.84 0.2497 6.67 0.3315
21 CR + HA 5.23 0.2486 2.78 0.2442 6.85 0.3407
Table 8: Results from all 4 Bora Bora water depth models as applied to each of the three
subgroups of samples. RMSE is given in meters. NRMSE is the RMSE normalized to the
depth ranges of the respective subgroup. NRMSE gives the error (on a scale of 0 – 1) of
the model. Model structure represents the variables that were included in each model (1 =
randomness model, CR = coral, and HA = Halimeda).
Water Depth (m)
Model #
Model
Structure
RMSE
(Platform-
wide)
NRMSE
(Platform-
wide)
RMSE
(Margin)
NRMSE
(Margin)
RMSE
(Interior)
NRMSE
(Interior)
0 1 11.73 0.3170 6.93 0.2772 7.18 0.2566
1 CR 9.58 0.2588 6.87 0.2748 4.93 0.1760
2 HA 10.25 0.2771 6.88 0.2753 7.11 0.2541
3 CR + HA 9.14 0.2471 6.86 0.2743 4.92 0.1758
Table 9: Results from all 22 Raivavae RDRR models as applied to each of the three
subgroups of samples. RMSE is given in meters. NRMSE is the RMSE normalized to the
ranges of RDRR measurements of the respective subgroup. The model structure
represents the variables that were included in each model (1 = randomness model, MU =
mud, SA = sand, GR = gravel, SO = sorting, CR = coral, and HA = Halimeda).
RDRR
Model # Model
Structure
RMSE
(Platform-wide)
NRMSE
(Platform-wide)
RMSE (Margin)
NRMSE (Margin)
RMSE (Interior)
NRMSE (Interior)
0 1 0.2044 0.2839 0.1884 0.3426 0.1916 0.3091
1 MU 0.1958 0.2720 0.1552 0.2823 0.1903 0.3069
2 SA 0.2037 0.2829 0.1541 0.2802 0.1911 0.3083
3 GR 0.2023 0.2809 0.1884 0.3425 0.1916 0.3090
4 SO 0.2044 0.2839 0.1739 0.3162 0.1911 0.3082
5 CR 0.1727 0.2399 0.1858 0.3379 0.1495 0.2411
6 HA 0.1798 0.2497 0.1869 0.3397 0.1545 0.2491
7 MU + SA 0.1942 0.2698 0.1536 0.2793 0.1902 0.3068
8 MU + GR 0.1946 0.2703 0.1515 0.2754 0.1903 0.3069
9 MU + SO 0.1947 0.2704 0.1506 0.2738 0.1903 0.3069
10 MU + CR 0.1602 0.2225 0.1511 0.2748 0.1492 0.2407
11 MU + HA 0.1610 0.2236 0.1546 0.2810 0.1531 0.2469
12 SA + GR 0.1956 0.2716 0.1521 0.2765 0.1887 0.3043
13 SA + SO 0.2033 0.2824 0.1506 0.2738 0.1910 0.3080
14 SA + CR 0.1680 0.2333 0.1493 0.2714 0.1491 0.2405
15 SA + HA 0.1680 0.2333 0.1534 0.2790 0.1522 0.2455
16 GR + SO 0.2020 0.2806 0.1714 0.3116 0.1908 0.3078
17 GR + CR 0.1725 0.2396 0.1858 0.3378 0.1489 0.2402
18 GR + HA 0.1798 0.2497 0.1865 0.3390 0.1532 0.2471
19 SO + CR 0.1695 0.2354 0.1611 0.2929 0.1376 0.2219
20 SO + HA 0.1777 0.2468 0.1737 0.3159 0.1530 0.2467
21 CR + HA 0.1725 0.2395 0.1758 0.3197 0.1473 0.2376
Table 10: Results from all 22 Tubuai RDRR models as applied to each of the three
subgroups of samples. RMSE is given in meters. NRMSE is the RMSE normalized to the
ranges of RDRR measurements of the respective subgroup. The model structure
represents the variables that were included in each model (1 = randomness model, MU =
mud, SA = sand, GR = gravel, SO = sorting, CR = coral, and HA = Halimeda).
RDRR
Model #
Model
Structure
RMSE
(Platform-
wide)
NRMSE
(Platform-
wide)
RMSE
(Margin)
NRMSE
(Margin)
RMSE
(Interior)
NRMSE
(Interior)
0 1 0.2582 0.3353 0.2825 0.3669 0.1154 0.3497
1 MU 0.2551 0.3313 0.2808 0.3647 0.0971 0.2942
2 SA 0.2550 0.3312 0.2809 0.3648 0.1047 0.3173
3 GR 0.2582 0.3353 0.2704 0.3512 0.0782 0.2370
4 SO 0.2480 0.3221 0.2754 0.3577 0.1114 0.3376
5 CR 0.2571 0.3339 0.2683 0.3484 0.0996 0.3018
6 HA 0.2582 0.3353 0.2632 0.3418 0.0941 0.2852
7 MU + SA 0.2547 0.3308 0.2806 0.3644 0.0748 0.2267
8 MU + GR 0.2546 0.3306 0.2686 0.3488 0.0748 0.2267
9 MU + SO 0.2429 0.3155 0.2718 0.3530 0.0970 0.2939
10 MU + CR 0.2546 0.3306 0.2682 0.3483 0.0863 0.2615
11 MU + HA 0.2547 0.3308 0.2619 0.3401 0.0885 0.2682
12 SA + GR 0.2547 0.3308 0.2689 0.3492 0.0748 0.2267
13 SA + SO 0.2396 0.3112 0.2716 0.3527 0.0772 0.2339
14 SA + CR 0.2547 0.3308 0.2682 0.3483 0.0751 0.2276
15 SA + HA 0.2543 0.3303 0.2620 0.3403 0.0445 0.1348
16 GR + SO 0.2474 0.3213 0.2556 0.3319 0.0648 0.1964
17 GR + CR 0.2570 0.3338 0.2597 0.3373 0.0458 0.1388
18 GR + HA 0.2582 0.3353 0.2431 0.3157 0.0404 0.1224
19 SO + CR 0.2446 0.3177 0.2566 0.3332 0.0973 0.2948
20 SO + HA 0.2473 0.3212 0.2579 0.3349 0.0939 0.2845
21 CR + HA 0.2556 0.3319 0.2605 0.3383 0.0941 0.2852
Table 11: Results from all 4 Bora Bora water RDRR as applied to each of the three
subgroups of samples. RMSE is given in meters. NRMSE is the RMSE normalized to the
ranges of RDRR measurements of the respective subgroup. The model structure
represents the variables that were included in each model (1 = randomness model, CR =
coral, and HA = Halimeda).
RDRR
Model # Model
Structure
RMSE
(Platform-wide)
NRMSE
(Platform-wide)
RMSE (Margin)
NRMSE (Margin)
RMSE (Interior)
NRMSE (Interior)
0 1 0.2382 0.2676 0.2076 0.3145 0.1827 0.2687
1 CR 0.2125 0.2388 0.1872 0.2836 0.1777 0.2613
2 HA 0.2311 0.2597 0.2055 0.3113 0.1823 0.2681
3 CR + HA 0.2123 0.2386 0.1679 0.2543 0.1762 0.2591